April 13, 2025 • 49 mins
In this episode, Joel Palathinkal chats with Albert Azut from Level VC about his transition from machine learning to venture capital. Albert shares insights from his time at Verizon and his shift into VC, discussing FinTech evolution, data companies, and tech team dynamics. They cover product management, data infrastructure, and the role of intuition in modeling, along with the impact of auto GPT and Chat GPT. Albert also talks about LP dashboards, venture fund evaluation, and diverse data use. The episode touches on venture fund specialization, flywheel effects, competitive strategies, and emerging sectors like precision medicine, deep tech, sustainability, advanced manufacturing, and biomanufacturing.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
For sure.
(00:00):
I mean, think I'll give you a little bit of ourmethodology, as you think will answer that
question.
And when we tackle when we looked at this spaceinitially, we said that we felt that it's first
of it's a very opaque market.
Yeah.
Welcome to The Investor, a podcast where I,Joel Palafinkel, your host, dives deep into the
(00:20):
minds of the world's most influentialinstitutional investors.
In each episode, we sit down with an investorto to hear about their journeys and how global
markets are driving capital allocation.
So join us on this journey as we explore theseinsights.
Okay, so we're live here.
(00:41):
I got a new guest, Albert Azut.
He's at Level VC.
So what I think which is really interestingabout their team and their platform is they're
using data and data science essentially to lookat funds and have a data driven approach in
their asset allocation strategy.
So that's why I thought I'd bring them on, youknow, because my community of fund managers and
(01:06):
LPs, we can just share notes live in real timewith the community and, and hopefully come back
with some interesting strategies across theboard.
But before we dive in, I want to introduceAlbert, he's the partner of the firm, and one
of the co founders as well.
So Albert, maybe you could just start welcometo the show, you know, excited about our new
(01:27):
friendship.
And maybe you can just start by just goingaround with your background and, and your early
career and how you got into fund to fundinvesting.
It sounds good.
Sounds good.
Well, thank you for having me.
Excited to be here.
Yeah.
So my my background, I'm I'm a softwareengineer.
So I guess, first and foremost, I went toschool in Boston.
And then, when I graduated, I went to work, onWall Street for about two and a half years as a
(01:49):
infrastructure engineer.
Mhmm.
You know, essentially doing hardcoreprogramming day to day.
And then I I left.
I was living in New York, I left to you know, Iwanted to start companies.
And so in New York, over the course of mycareer, I started a few companies in different
areas, all technical, some in in ad tech.
And then the last company I did was was focusedon machine learning operations.
(02:14):
So we were quite early in in the machinelearning movement, when it sort of, helping
enterprises move from raw data to machinelearning at scale.
And we were, you know, an early adopter of a ofa project called Apache Spark, and we
contributed to it.
And, essentially, what we did was we builtpipelining systems, for enterprises, which then
became the category ML Ops.
(02:35):
Mhmm.
And so we built that company, in New York.
I had a, you know, really strong team, and weended up selling the company to, Verizon in
2015 via via an AOL.
Mhmm.
And and that was an amazing process, difficultbut amazing process.
And as part of that transaction, you know, I Ihad the opportunity to to leave New York and
(02:55):
and go to California.
Mhmm.
So in in 2016, I picked up and I ended upmoving to to Palo Alto, California
Mhmm.
Where, you know, I worked for Verizon, forabout a year and a half.
In in Palo Alto, they have a, you know, anapply a research group applied research group
where, you know, the team works on, you know,very large scale machine learning problems
(03:17):
Mhmm.
Tackling very you know, sort of variousbusiness initiatives, you know, within AOL and
Verizon.
Everything from, you know, understandingwhether two devices are the same user to, you
know, predicting, you know, will somebodyconvert on a piece of advertising?
And all those are very, very hard problems.
So during that that time, I I was able toreally get a lot of experience in, you know,
(03:39):
scaling very large Mhmm.
Machine learning systems.
And then and then I started to get prettyheavily into deep learning and what became sort
of modern machine learning, what we see today.
And, you know, at that point, I I actuallyinteracted with one of the investors in my
prior company, and he was his name is Bobby.
He was putting together a new firm, in venturecapital, which which I believe had a very
(04:03):
interesting strategy, still does, which wasinvesting multistage, with a focus on
enterprise software.
So it's a firm called Coda Capital in SanFrancisco.
Basically, you know, they would invest earlystage, growth stage, and also public.
Yeah.
And that was an interesting purview on thewhole landscape because as you're investing in
private, you're getting sort of perspective ofpublic positions.
(04:27):
And so I ended up joining the firm, you know,as a partner in in 2016, and I stayed there for
four years, you know, helping build the firm.
And I was focused on the private investingside.
So all of the early stage investing and, youknow, my area of focus or at least, you know,
where I where I had the most appeal was in, youknow, deeper tech Mhmm.
(04:49):
Anywhere it's like semiconductor all the way toinfrastructure.
And then we did some application level thingsas well.
But that was really my my focus and and sort ofwhere I where I like to to spend time.
And then I did do a little bit more lifesciences, had more life sciences exposure
during the end of of that.
So I was there four years, and then duringduring COVID, ended up, you know, coming to
Miami, where this is where I'm originally from.
(05:11):
And Yeah.
We we we just were really happy here, and weended up just ended up staying.
That's Sure.
That's the short end of the story.
And in that, I had the opportunity to, youknow, transition out of CODA, which was an
amazing experience, but, you know, think aboutwhat I wanted to build.
And so that's how I ended up, you know, atbuilding level ventures.
Yeah, I see that often, I see a lot of buildersthat became VCs, including myself that are
(05:36):
starting to kind of build some data drivensoftware that you know, what happened with me
essentially was I became an I was in tech,worked in venture and still doing still working
in venture, but also starting to slowly build acouple products to help my workflows and then
just sharing it with the community and seeingif they they think it's useful.
If it's not, you know, maybe just iterating offof that.
(05:58):
And one of my favorite people is Yo, hey, Idon't know if you watch Yo, hey, on on Twitter,
he is he's at untapped ventures, and he buildslike a bunch of chat GTP, chat GPT algos like
in in real time on on Twitter.
So he he built like really cool, like, techparsers and and all that stuff.
(06:19):
So I think it's really cool.
And his his following just recently got a hugesurge because, you know, the content that he's
sharing, it's definitely it's definitely ofinterest to the community now.
Right?
People are people are getting into promptengineering.
So there's a lot to unpack and nerd out on.
You know, one of the first things is, you know,your your early days working on Wall Street
(06:42):
because I live in New York still, I've beenhere, you know, kind of like the way that you
think about Miami.
That's my family in New York City, you know, sowe live here, we've stayed here for about a
decade.
Tell me about the ecosystem when you wereworking on Wall Street and kind of like how
FinTech was developing like FinTech and data,data companies, because it's way different now,
(07:02):
right?
It's very vibrant.
So I'm just curious to know and I can tell youabout like how it was when I was there.
I worked at FactSet, you know, building a lotof their enterprise FinTech.
So it was kind of slowly developing, but lovelove some insight in terms of, how it was when
you were there.
Yeah.
I mean, it's a long time ago, so I'm I'm datingmyself.
But, you know, and I was I was really involvedin in sort of, like, core infrastructure
(07:25):
porting.
You know, I think if anything, can say that thecertainly, the architecture, you know, like,
sort of your traditional monolithic softwareapp and how you deploy software Mhmm.
And the speed and velocity at which you deploy,you know, it's just it's a whole different
world now.
You know?
The process from by which you develop software,you check it, you deploy it into this sort of
(07:46):
very monolithic process.
You know?
It's so, you know, sort of highly sort ofbureaucratic and regulated.
Yeah.
That was sort of one major yeah.
One that's a huge change.
Obviously, that's a consequence that's justchanged.
The time curve has changed a lot.
And then, you know, when it came toquantitative methods, which we did I did a
little bit of that while I was there.
Mhmm.
And I was I was getting sort of very interestedin, you know, quantitative finance and and
(08:09):
thinking about time series analysis.
But those methods, you know, are sort of verydifferent, than what you see today.
I mean, a lot of the same primitives are there,but Mhmm.
The world has changed in terms of how you howyou sort tackle those kinds of problems at
large scale.
One big thing I would say in terms of techteam, back in the day, the business pretty much
had all the power.
They're like, hey, you need to build this bythis quarter.
(08:32):
And then it's really like a project.
It's like, hey, this needs to be done by q1,like all these back end scripts or, you know,
the the platforms that you were building.
I feel like now the engineers have a huge stakeat the table.
And there's also discipline of productmanagement.
I'm assuming that they didn't have, like, theproduct management discipline back then.
(08:52):
So it's a bunch of like so was it like thebusiness stakeholders?
It's like, look, know, you need to or salesthat's like, hey, you need to launch this by
two months because the customer wants it.
Yeah.
Yeah.
I mean, it was it was basically the CTO makingdecisions and Yeah.
Down.
You know, I guess it was more of an IT functionand less of a strategic technology function.
Yeah.
I think it's been the major shift Mhmm.
(09:13):
Which is software is sort of the business in inmany cases.
And and back then, it was just sort of asupporting infrastructure for many of the other
things that the business was doing.
Some of it was critical, mission criticalinfrastructure, but certainly sort of forefront
in terms of the business itself.
And I think that obviously has changed, youknow, drastically.
Yeah.
And with that, I think the the organizationalstructure around, you know, product management
(09:37):
and thinking about engineering process, butalso just developers themselves are you know,
they have the ability to choose various toolsand tooling and infrastructure.
And since we're on the cloud, there's so manythere's such a combinatorial explosion of
things you can do to get software deployed.
Yeah.
Power has certainly shifted, you know, over thelast, like, twenty years or so.
Yeah.
No.
Absolutely.
I mean, the engineers have a huge influence.
(10:00):
They're not just kind of taking orders, theyactually, I think it really has to be an
aligned strategy with design, product marketingengineering.
And I think sometimes like the roadmaps need tobe aligned too because I mean, the engineers I
work with, know, they they wanted to continuebuilding interesting infrastructure and you
(10:22):
know, make the platform faster where like, youknow, I was a product person, right?
So I was more excited about adding new dropdowns or like new, new graphs and charts and
analytics.
And, you know, trying to get I think part ofthe product managers discipline is like,
getting the engineers bought sometimes justhelping them understand.
In my experience, like I burned the mostbridges.
(10:43):
Well, I haven't personally on the record, butbut like, I think engineers in general get
upset, like if there's throwaway code, right?
If they build something, and then it actuallycame to of no use to the customer and then they
have to like rebuild it again.
So I think in my experience, it's been and Ilove to hear your input on this, but like the
modern approach has been essentially, you know,thinking about the bigger vision of the
(11:06):
company.
So if you are going to go on mobile, have theengineers know because they could probably
choose the right infrastructure or like theright, you know, code and tools to kind of do
that where if you know you're never never goingto be on mobile, then there's probably a
better, you know, infrastructure to use or or,developer methodology to deploy if you wanna
(11:28):
achieve that goal.
And I I I agree.
I think if if know the endpoint on sort of, youknow, how you're servicing customers and
whatever they're gonna be on, I think it helpsmake decisions
Yeah.
A priori before you start building.
But at the same time, you know, there areplatform shifts.
Right?
And and Yeah.
You know, something like, you know, like LLMsor foundational models, I think those things
require sort of a rethinking and a re you know,sort of complete paradigm shift and design
(11:52):
patterns of how you build stuff.
So so I think, like, you can you know, as aslong as the teams are agile and they can sort
of there's not a lot of technical debt built upin the system, and they're able to sort of
quickly abandon things and, and sort of moveinto infrastructure that's more efficient, you
know, and and sort of and that's why I thinkthe breaking down of the monolithic app was a
big thing because
Mhmm.
(12:12):
It pulls you to, you know, very be more agileYeah.
Through how you develop.
But, yeah, I think those are sort of bestpractices that Mhmm.
You've learned, like, this learning rate thathumanity has had is, you know, in terms
of Yeah.
Software more efficiently.
And we're still really at the beginning of thatas well.
Yeah, no, absolutely.
And then, you know, I think it would beinteresting for the audience to to hear your
(12:35):
process in terms of building one of your youknow, I know you launched like four companies,
but maybe you can tell me, maybe the first oneor maybe, like, one that you choose would be
better to share, but essentially how you cameup with the idea and then how you got it to
market and
Yeah.
And
how you kinda scaled it.
Yeah.
I guess maybe this the last one is isappropriate.
(12:56):
Mhmm.
Yeah.
You know, I think the I guess when you're veryearly in an industry, and you're thinking about
problems that the market not necessarily isthinking of, so you you really have to, like,
pace yourself and, you know, have a a go tomarket motion and a sort of a product market
fit that needs to evolve over time.
Yeah.
As we're sort of recognize the problem becauseI think one of the dangers is like having to
(13:17):
teach customers.
Mhmm.
Well, it sort of creates a lot more friction.
Yeah.
So I think one of the things we learned thereis is when it comes to product management is
really just being on time in terms of the thesort of the problem set that you're that you're
solving.
But I think in building in building the productand iterating on it, you know, we had some
ideas as to, you know, the core technology thatwe wanted to build.
(13:41):
We felt that, you know, when you're dealingwith very, very large datasets and you're
trying to do something like, for example, youknow, unsupervised clustering, it's very, very
difficult as that dataset grows, and you needto have a methodology that's distributed that,
you know, you can sort of solve the problemefficiently even even with this, I guess,
you're, like, halfway there.
(14:01):
Mhmm.
So that's sort of where we focused, you know,our our efforts.
And so we had a really good core technology,but then there was a gap with the actual
product and how do you how do you into aninfrastructure.
So that's that was kind of a learning process.
Like, how do you Mhmm.
You know, architect a product that has that,you know, fits within a particular customer
workflow.
Mhmm.
And how do you go about selling that?
(14:23):
You know, you know, what's the productmarketing to it?
So that was a big big you know, and sometimesyou have to rework the whole product because
you just, you know, you UX based, and you justdidn't have the right fit.
So that was a very long process and a longlearning that we had there.
I think the outcome was good, but we werecertainly early in in the market.
Yeah.
Times is is is a disservice, you know, becauseyou need to be able to survive as well.
(14:47):
So Mhmm.
That was a big learning for me.
And also just building large scale machinelearning systems, I think, is is completely
different modality than, you know, than doingsomething that's just sort of traditional
software.
Yeah.
I mean, I'm sure the cost has definitely gonedown essentially with with usage based pricing.
(15:07):
But can you tell us a little bit about just howinfrastructure has evolved?
Like in terms, you know, because right now,when I think of data platforms, I know Google
has like Google compute, and there's likedigital ocean.
But if you're trying to build kind of a largescale data system, what do you need to think
about if you're like a founder?
(15:28):
Yeah.
And I think there's, you know, I guess there'sa few pieces of it.
And then like Hadoop was the thing that peoplewere talking about for a long time, you know,
distributed systems, but that was like, I, Iwas kind of in the space, like back in 2014.
So a lot of those buzzwords, like big data andstuff, I'm sure, like, a lot of that has
changed.
Oh, yeah.
I don't even hear I don't hear it as muchanymore.
(15:48):
Yeah.
I mean, Hadoop was an interesting solution forthe for the period of time.
I I I think it's it's much more overhead thanthan what's needed to manage data.
We see that with sort of flat data lakes.
Yeah.
More you don't need that sort of overhead.
It's more understanding the protocols and andsort of having the ETL piece of it.
But, yeah, I think all that wholeinfrastructure has changed.
And and, like, Hadoop was really meant toserve, like, the original distributed computing
(16:12):
framework, which is like a Macrius.
Yeah.
Came out of Google.
And, you know, that was the beginnings ofthinking about distributed computing in large
scale systems.
And I I really think there's a lot of the sameprimitives ended up being in in sort of some of
the deep learning techniques for for trainingthat we see today, obviously, with has its own
nuances.
So I I think that that certainly evolved.
(16:34):
But if if you're if you're a team building,infrastructure, data infrastructure, I think
you have to think about the, you know, thenature of the data.
Right?
Unstructured, structured, and and sort of howthat's stored and, you know, what are the
what's the schema for storage and what are thesystems for storage.
And then really, the data pipelining piece, Ithink, is so critical.
Mhmm.
A lot of the value I think you get besides, thedownstream modeling, which I think, you know,
(16:59):
you it's it's nuanced, but a lot of the valueyou get is really building data infrastructure
that is scalable, that's able to manage andintegrate datasets.
You know, for example, one big problem that wehave, as you can imagine, is is entity
resolution at level, right, which is, like, youhave many different datasets that are coming
in.
You have different entities in the dataset,like companies, people, etcetera, and it's very
(17:23):
noisy.
Right?
So you you're not you're not always surethere's ambiguity, but it's also very noisy.
So you have to be able to resolve a transactionor an entity or a company into the same thing
Mhmm.
With high, you know, high integrity.
And that is that's sort of a data engineeringproblem.
Right?
Sure.
At our scale.
So those are the kinds of things that you needto do very well and prepare the data in a way
(17:45):
that then could be used for modeling veryefficiently.
So I really think, like, a lot of the gruntwork actually, I think a lot of the grunt work
moving moving forward will also be very datainfrastructure heavy versus modeling work,
which I think will be sort of out of the boxreally strong.
So, anyways, I think that's, like, the keything is setting up
the
right the right pattern, the rightinfrastructure, the right visibility
(18:07):
observability in your data infrastructure.
And then from there, you know, I think is isreally, having I always feel like it's
important to have an intuition on what you'retrying to model.
Mhmm.
You know, you can you can apply, like,traditional out of the box stuff and say this
is you know, here's a target variable.
Let me try
to Yeah.
Solve it.
But I think you need to have an intuition onwhat the nature of the data that's being
created is, how it's being created What'sobservable about it, what's not observable
(18:32):
about it, and then model it based on thatintuition.
And then from there, you can choose models thatmight suit the purpose.
Okay.
And so I think that's something I've kind oflearned over time.
And not everybody has an intuition for howthings how they generate.
It's not easy.
But, you know, we've always that's the that'skind of like the modality and the the paradigm
that we've always operated under.
And then let's talk about kind of the new powerparadigm shift with, you know, auto GPT and
(18:57):
chat GPT.
Like, what what's kinda like the infrastructurethat needs to change?
I guess, are we are they still using datalakes?
And then and then I guess what's the humanimplication?
You know, how is this gonna impact maybe ourindustry?
You know?
I mean, can we can we ask prompts to look atcertain parameters on fund performance and then
(19:18):
have that pull up?
And is that kinda what you guys are thinkingabout with your, you know, your kinda data
driven approach and your data sciencebackground?
Well, I I think it's, it's certainly a aplatform shift, a significant platform shift.
I think the way we design applications Mhmm.
Moving forward is gonna change drastically.
If you just look at some of the primitivesaround you know, some of the things we're we're
(19:40):
talking about just today, like promptengineering.
Mhmm.
You know, it's like, that's a term that, youknow, really is, you know, just completely new.
Yeah.
Human or few shot learning or things like thatwhere you really have to think about, you know,
how do you build a system leveraging thesecapabilities in a way that provides value.
I think the way you build applications is nowgonna be completely different.
(20:00):
And you can also assume that asymptotically,you know, the systems will get a lot more a lot
more powerful.
The context window will get larger Mhmm.
And get cheaper.
You know, you're gonna be on a on a veryrapidly diminishing cost curve.
And so I think that's gonna make a lot oftraditional software development and
applications irrelevant Mhmm.
In many ways.
And I think you'll have a situation where,yeah, incumbents will integrate a lot of this
(20:23):
into their core application, and then someincumbents will no longer be relevant because
of we retrieve information, you know, how wecreate content and media and all that will
completely completely change.
Yeah.
That being said, I think it's you have to sortof take it with a grain of salt as well.
I mean, there it's not always good to, like,use, you know, historical, you know, sort of
(20:46):
corpus to generate future context.
Like Mhmm.
You know, you're gonna get these things likehallucinations and, you know, you're gonna get
sort of data integrity issues.
And at the end of the day, you know, the corpuson which it's training on is the human corpus,
which needs to evolve.
Right?
So Mhmm.
So we we, at least currently in its currentmanifestation, the way we're using it is more
(21:07):
to be able to embed entities into some sort oftopic space.
Mhmm.
So if I look at a company, I wanna understandvery, very low level, what is it doing?
Not not is it in health care, but is it doing,you know, DNA synthesis using enzymatic
approaches.
Right?
That's the kind of, like, understanding that wewanna have around companies and then founder
(21:29):
expertise and also company expertise and andthen also, you know, manager expertise, fund
manager expertise.
So we use it more for understanding the spaceof topics in which Mhmm.
Businesses are operating in.
Yeah.
At a very fine grained resolution.
I think we could eventually build a system thatcan integrate into our workflow so we can have
(21:50):
textual questions and responses, know, based onour own system of data.
So you you'd wanna combine your, you know, sortof large foundation models with, let's say,
your own internal datasets and knowledge Mhmm.
That you can actually, you know, extractknowledge from your own system in a much more
seamless way.
And then maybe, like, tie threads betweenthings you wouldn't have otherwise I've seen.
(22:11):
But I think out of the box for what we'redoing, we don't we don't you know, we're not
sure that it's the right approach to be ableto, like, make decisions.
Yeah.
That that's our current, you know, thinkingright now.
Well and I guess you can use, you know, us bothbeing from Wall Street.
You know, there's live stream data that you canpull in from, like, exchanges.
Obviously, we don't have any exchanges, but wehave data feeds.
(22:33):
So if we were to kinda build something thatcould be valuable for LPs, what do you think
they would want to see on their dashboard?
Like, if they're if they're like someone justtrying to do research on, hey, which which is
the best one?
Because there's so many of them.
A lot of the standard performance metricsthat's that benchmark, that's pretty much table
(22:53):
stakes, right?
So whether, you know, you're looking for acertain level of net IRR or a certain level of
ROIC and even an industry, right?
There's a ton of it.
There's a ton of funds focusing on AI.
There's a bunch of funds focusing on healthcare.
They all have the performance.
They're all great.
Right?
So so what else could we pull?
And, you know, if someone were build someonewere to build some type of data platform, how
(23:17):
could you determine kinda what is good?
And I guess Yeah.
Yeah.
For sure.
I mean, I think maybe I'll give you a littlebit of our methodology.
I do think we'll we'll answer that question.
And when we tackle when we looked at this spaceinitially, we said that we felt that it's first
of it's a very opaque market.
Yeah.
And there's a lot that's unobserved in themarket.
Right?
Mhmm.
And I like that.
(23:37):
That's an an opportunity we like, right, whichis a difficult one.
And then there's probably three things that aretrue, about venture specifically just given the
underlying, like, mechanics returndistributions.
One is that, you know, manager, performance ishighly dispersed.
Mhmm.
Right?
So that's it's traditionally, like, that's thecase where you have a huge dispersion in in
performance.
(23:57):
So if you're not really in the top quartile ofa fund manager, you know, you might
underperform the whole market, even even otherasset classes, which we've seen.
Yeah.
The second is that sequential correlation is isrelatively low.
So just because you did well in one period oftime doesn't mean you'll do well in the next
period of time.
To some some extent, there is some correlation,but it's not it's not extremely strong.
(24:18):
And the third thing is that it's very timedependent, you know, as we know, vintage and
cycle and background market conditions reallyaffect returns.
You have a lot of portfolios today where, youknow, they look really good.
But, you know, over the next few quarters asthese sort of 400 unicorns that have not raised
since 2021, you know, what happens to thoseportfolios, right?
And so we always had the intuition that thetrack record was certainly not enough.
(24:44):
Sure.
And it's an it's an indicator, but it's notsort of a strong indicator.
And so when we started to look at the market,we realized that, you know, one of the things
that, you know, what are the stable things thatdrive, you know, durable performance in any
market like this that's that's has a lot ofuncertainty.
And what we realized, yeah, a lot of it isnetworks, right, is is understanding the
relationships between entities, the flow ofinformation between them.
(25:08):
And if if one, you know, individual influencesanother person's decision, that actually is a
very strong signal.
And so we've taken the approach of trying toreconstruct, the network of behavior and
activity using what is basically raw data thatdoesn't have any network structure.
And so that's what we do.
(25:30):
U we construct features that are focused on,network, network durability, recency, strength,
etcetera.
And we use those features to then makepredictions on on future performance calibrated
against historical past.
The benefit of doing that is it's it'sinvariant to time.
(25:50):
Mhmm.
You don't get a you don't get a lot of thesesort of false positive effects of of purely
performance based metrics.
Yeah.
You actually realize the underlying driver isthe asymmetric access that some have in the
market, and we think that that's a big driverof that network.
So to answer your question, understanding theco investor base, understanding the sourcing,
(26:11):
the, like, the the discrete sourcing pipelines,the founder networks that they have, the those
kinds of things, I think, are so much moreimportant than looking at just, performance on
its own Yeah.
Which is just a side effect of those things.
No.
Absolutely.
I mean, I think that's also probably directlybeen related to some, you know, fundraising as
(26:35):
well.
Right?
And people that have been able closeallocations.
Obviously, there's regulations in terms of whatyou're allowed to share publicly, but just
people that are creating a community publiclybecause there's education that they're putting
out there.
Like if you're like the guy Yohei, I'lldefinitely ping his his Twitter.
You know, he posts a lot of content on AI, andthat's just built an engaged community.
(26:59):
And, you know, especially on Twitter that, youknow, there's a lot there's a huge engaged
community in the VC space there.
So I think using some of those social networks,I think, could help.
And is that what you're referring to?
Like, maybe some of the the data that you seeon Twitter, and then maybe also on LinkedIn?
And I mean, TikTok is up and coming too.
There's a few, you know, a few fun managersthat are really, really using that.
(27:23):
So I don't know if that's if that any of those,like, social networks are helpful as well
beyond Yeah.
I think it's a signal.
I think it's a signal for yeah.
Certainly, a fund has a flywheel aroundcommunity Yeah.
They're creating a lot of content, and there'ssort of this flow of of talent right now, I
think it's a very strong thing.
Yeah.
Although we we focus less on virality and moreon network, which I think is actually a
(27:43):
distinction between the two.
Because you could probably get information veryrapidly, but doesn't necessarily mean you have
durable network relationships.
Yeah.
So we
balance both of those things, but what I what Iwhat we're referring to more is very strong,
for example, co investor networks that that arehighly durable, let's say, or or access to
founder networks that are durable.
Mhmm.
It's not sort of one shot kind of things whereit really is.
(28:05):
Yeah.
Yeah.
That's what I'm talking about.
Okay.
And a proxy for that could be social because ifyou do have a big presence, then wanna be
working with you.
So that's important as well.
Yeah.
And you could probably find some well, I guesssome of that data would have to be probably
volunteered to, to be provided as they'reapplying to, you know, your your your platform.
(28:29):
And then some of that stuff you could probably,you know, just get on the Internet, you know,
just through data datasets and stuff like that,I'm assuming.
Right?
Yeah.
Yeah.
We we've always taken the approach that, youknow, from an evolutionary perspective, there's
gonna be more data, not less.
And especially in the private markets whereit's been very opaque.
There's more accessibility and democratizationof information.
(28:49):
Mhmm.
And so then what matters is not so much thedata itself, but, you know, sort of how you
manipulate it and insights are.
And and actually think networking the data isis a big, import it's an important element,
right, which is to take disparate datasets thatare orthogonal and put them together in a way
that's very unique.
I think that, like, that sort of thinking, Ithink, is more important than just getting
(29:13):
access to, let's say, Crunchbase or PitchBook,which everybody can.
Yeah.
And I'm saying that I think, you know, yourunique data science strategy to your point,
right, taking those sets and stitching it tocreate a new type of, maybe your own
performance score.
Right?
I mean, that could be really, reallyinteresting too.
It's kind of a novel because there isn't beyondthe traditional metrics that we've seen for
(29:34):
years.
I think there could be a network score or likea network performance score based on what you
guys have been, developing.
That's, that's really interesting.
And, you know, on top of that, you know, on topof the network, what's your, you know, for the
for the emerging managers that are applyingthat fit your criteria, you know, what are you
(29:54):
looking for in in managers and and what aresome of the things that managers should think
about when they're applying to, you know, yourplatform as well?
Yeah.
I mean, I think this we have an opinion on onthings that we look for.
Mhmm.
I guess there's a few different areas.
You know, one obviously is fund size.
You know?
Mhmm.
We're very strong believers that venture doesnot scale very well and that smaller fund sizes
(30:19):
have have big potential to outperform justrelative to the fact that, you know, you you
can return multiples of the fund with one exit.
Yeah.
And so we do look for funds that are reasonablysized.
Mhmm.
And and that's important.
And it's also an area which is very hard to todiligence.
Right?
Because Mhmm.
Usually, they have unless they stay purposelysmall, typically, they don't have a lot of
(30:40):
track, you know, to to Mhmm.
On, and it's more opaque.
Yeah.
And it's also underfunded.
So I think, like, that's one thing we look foris the right portfolio construction strategy,
including funds.
The other thing we we really like, more andmore is just special you know, to some extent
specialization.
Yeah.
We we believe that technology is getting morecomplex, and the the nature of the problems
(31:03):
that we're gonna be solving are more are verytranslational.
Meaning,
like, core science, either physics or biology,or even AI.
So these are deeper areas and translating theminto a business.
Mhmm.
And that we believe requires, you know, deepdomain expertise and and more so the ability to
attract entrepreneurs that are building thesemore complicated businesses.
(31:25):
Obviously, they wanna work with those that havehad experience in that area or understand their
businesses.
So we actually think specialization is isimportant.
And, like, it could be, like, domainspecialization.
It could be something else.
Right?
Yeah.
For the market specialization.
But so we do look for specialization generally.
Mhmm.
Especially in environment like today where it'syou just can't invest and expect good results.
(31:46):
Yeah.
So that's another thing we look for.
I would say flywheel, it's just hard toquantify.
Mhmm.
But there's something about, you know, like,every success and every investment you make
should have nonlinear effects on nextinvestment.
Right?
So it's not like you're just making randombets.
You're there's a, you know, a community angleor Mhmm.
(32:07):
An expertise angle or some other angle that'sbeing built over time, which is which is gonna
give you more results and nonlinear outcomes.
So we we look for that.
It's hard it's hard to quantify what that is.
Mhmm.
And then the last thing is just people.
You you wanna have sort of very intellectuallyhonest people that understand, you know, where
(32:29):
conviction can fail Yeah.
And sort of where you need to have diversity ofof investing and where you have to think about,
you know, how you treat entrepreneurs and howyou treat your investors and and how you, you
know, sort of how active you are or just activeyou are.
Just there has to be a thoughtful approach towhy you're doing things.
Yeah.
Anyways, those are just some of the things thatthat we look for.
(32:49):
In addition, we have we use our our technologyvery deeply, in our diligence to, you know, us
understand different features of the managerand, you know, based on their prior investments
and network and all that.
Art and art and science, I would say.
Mhmm.
No.
That's that's awesome to be building and also,using the tools yourself.
(33:10):
When it come I'd like to just double down alittle more on the portfolio construction.
So, you said reasonable size.
So what what what would be reasonable and thenkind of hit the board boundary of,
unreasonable?
And then I'd love to kinda have you double,talk a little more about portfolio construction
and your thoughts on that.
(33:30):
Yeah.
I think it's it's there's not a rule in youknow, there's not, like, a rule of thumb around
around these things.
I I think you wanna, you know, and it alsodepends on ownership, you know, like, your
ownership targets.
You know, I think there's really, at least inthe seed, there's really two, I think, two
kinds of investors or two two stable points ofinvestors.
One is sort of your positive sum game investorswhere, you know, they can fit into a round.
(33:51):
You know, they maybe they'll put quartermillion or half a million, and that's you're
not you're not sort of pushing anybody out, butyou could get on really, really strong cap
tables and that gives you a real return.
And then you have those that that are that aremore competitive and zero sum game where they
need to, you know, they need to deploy3,000,000 in a seat, and that that means that
others cannot.
And so those are the kinds of the two types.
Mhmm.
I actually think they require differentportfolio construction strategies for sure.
(34:14):
Mhmm.
I think for us, like, we the smaller, thebetter, to be honest.
Like, we we think, like, $5,025,000,000 is areally good fund if you have three companies.
Mhmm.
Where where you have, like, a relatively goodcost basis for ownership.
You you could manage a hundred and $50,000,000fund with with a good amount of, let's say,
fifteen percent first ownership.
(34:35):
Mhmm.
And let's say, a thirty thirty companies withwith with some follow on reserves Yeah.
Which I think is another really good state.
We call it a core position versus a singlemarket.
So I think those are, like, the two median sizefor us.
Like, our median fund size is a hundred millionin our portfolio.
We have 20 funds.
It's a hundred million.
Median.
I think that's the right that's the right sizeif you care about, you know, carry.
(34:58):
Right?
If you care about incentives, generallyspeaking.
So Yeah.
We we've done a lot of statistical analysisanalysis on this.
I mean, just on the the distribution of returnsand venture, and how you you can sample, you
know, a set of companies from a portfolio likeyou really wanna be in that in that size range,
I think.
Mhmm.
And, can you share a couple examples of a goodportfolio construction for a positive sum game
(35:23):
strategy versus a competitive zero sum gamestrategy?
Yeah.
I we we like so if you have, let's say, a50,000,000 or 25 to 50,000,000, that's like a
nano.
But let's say you have a $50,000,000 fund.
Yeah.
Let's do 50.
Let's do maybe like 50 for both, maybe to keepit simple.
Right?
Or 50.
50.
Yeah.
So 50 for a positive sum game, I guess.
(35:44):
How many companies?
How much reserve?
How much ownership?
Yeah.
You know, we we care about reserves to somedegree.
Mhmm.
I I don't I wouldn't we're not, like, sort ofoverexposed on reserves.
We actually have more companies.
Yeah.
But we we think that if you're sampling 30 to40 I say sampling, obviously, you're investing.
But if you're investing in 30 to four 30 to 40companies in a portfolio with, let's say, two
(36:07):
to 5% ownership, you know, then if you havesort of a very if you have a few sort of
nonlinear outcomes, I think that returnsmultiples Mhmm.
The portfolio.
Yeah.
And if you especially if you're recycling.
So, that's that's what we think.
So I I guess what I'm saying is, 50,000,000overly concentrated portfolio, let's say you
have 10 names Mhmm.
Makes us quite nervous.
(36:29):
Just because it's it's just you don't know.
This is a very uncertain environment.
Yeah.
It's it's, like, structurally uncertain.
It's not like we don't know.
It means that we we don't know what we don'tknow.
Yeah.
And so you there's no way to a priori know whatwill be successful, you know, no matter what.
And so Mhmm.
You have diversity in position sizing anddiversity in number of positions in your
(36:52):
portfolio, in order to be able to get thesesort of nonlinear outcomes.
You know?
You could get lucky, but that's not a strategy.
Yeah.
So that's what we think on, like, the$50,000,000, you know, positive sum game.
And then, you know, like, some of the firms wementioned before we started the the call, I
mean, that's that's what they do.
Mhmm.
And what what about the deployment schedule?
Are they looking to deploy most of the capitalin, the first three to four years and then and
(37:17):
then try to think about their next fund?
Or is it are they actually still deployingthroughout the whole the fund?
No.
They they typically have a three yearinvestment period.
Mhmm.
Yeah.
Especially for the smaller funds too.
They wanna get
yeah.
And you wanna get time diversity.
Right?
Because Mhmm.
You you know, a lot of, like, for example,like, you know, we know that LLMs are like a
(37:37):
platform shift.
We don't you know, I of course, we don't knowasymptotically what what the market structure
will look like in terms of value.
Mhmm.
But you wanna be able to take a few bets,whether in the infrastructure side or tooling
side or even, you know, sort of verticalizedapplications.
Yeah.
You really you know, you wanna be able to dothat over time, and you couldn't have done that
last year, and now maybe this a lot more.
So time diversity allows you to get exposure todifferent themes, thematic areas.
(38:02):
Mhmm.
And, also, you don't you're not sort of stuckon one vintage from a valuation perspective.
So I think you wanna have diversity.
Okay.
A lot of the great managers, I think, over thelast three years when things were very loose,
were the ones that paced properly.
You know, they were very, very thoughtful invaluations, and what they paid and how they
(38:23):
paced.
And so a lot of to be honest, a lot of ourmanagers haven't been deployed that much.
Yeah.
Our fund, you know, 10% has been deployed.
Mhmm.
You know, we're a fund of funds.
So Sure.
Which means that the managers aren't callingthat fast.
Yeah.
Which is why we like we like to see.
Mhmm.
So, yeah, that's that's what I would say aboutthat.
Okay.
And then the competitive who's here some game,how to I'd be curious to see yeah.
(38:47):
Yeah.
I'd be curious to see how that differs from I'massuming just a number of companies, I guess.
Or?
Yeah.
I think it's a it's probably a it's a largerfund size.
Think Yep.
You wanna be able to take larger stakes, youknow, in the company.
Okay.
So this will probably be maybe a 75,000,000 ifwe did the example here.
Yeah.
I'll say a hundred 75 to a hundred and likethat.
And there there you wanna have, you know,you're gonna have larger checks.
(39:09):
Right?
And you need to have some reserves.
I mean, a founder's gonna wanna make sure youhave reserves.
Right?
Mhmm.
They're signaling and other things.
And so it just requires a bigger fund.
And then you're competing with with you know,you need to have a team, I think.
Right?
And management fees matter, and you need tohave, a team of people around you to support
the companies because the founders, you know,that's you know, it's really good founders
(39:30):
understand who are the right investors.
Right?
Mhmm.
So so I think you need to have a biggerstructure in order to be able to do that.
Yeah.
And they're like, we never sort of vary from 20to 35 companies in a portfolio.
I think anything less is way over concentrated.
Sure.
It's it's very it's just too risky.
So yeah.
So that's what I would say.
And plus sometimes there's there's good extent,like, seat extensions where you can take more
(39:52):
ownership, and so you wanna have reservesthere.
And then Yeah.
Pay.
So I think there's 30, you know, 30 companies,let's say, twenty five, thirty companies,
hundred and 200, hundred and 20 five, you know,million dollar fund size Mhmm.
With, let's say, 50% maybe reserves
Yeah.
Is, like, what you typically see.
And think that could that could be a very goodstrategy if if you have conviction that they
(40:14):
can lead rounds.
Okay.
Yeah.
They that they can actually take the lead.
That's good.
That's good insight.
And then have you guys ever thought about doingsome direct investments along with your
strategy?
Because obviously, that's one of the biggestperks with some of the emerging managers and
just strategic funds.
(40:34):
You know, they they always leave some coinvestment opportunities.
So, you know, I guess, is that something youconsidered?
And then I've seen a lot of funds now slowlyhave, you know, kind of an LP, you know, on the
other, we'll we'll address this question first,but I've seen some funds also have kind of an
LP well.
They're like, you know, they mainly invest indirect companies, but they have a small carve
(40:57):
out to invest in some emerging managers too,which I think is great for for access to deal
flow and just networks that they normally don'thave.
Yeah.
Yeah.
So we we do a co invest, actually.
Mhmm.
We do a lot of that.
Cool.
It's always been our strategy to to support theunderlying portfolio companies later stage.
So, typically, that that happens with a seriesb or series c Mhmm.
(41:20):
Where the the size of the rounds are largeenough where we can, you know, come in
Yeah.
And and piggyback either on the relationship orthe pro rata of the manager.
And we share economics with our managers veryopenly.
So we do that all the time.
Mhmm.
And sometimes there's a series a extensionwhere we we we, like, we think the company is
really strong and just needs more time.
Yeah.
We're happy to, like, support that situation aswell.
(41:42):
Mhmm.
There's, like, asymmetry of information.
So, yeah, we we and, you know, where we focusis really an enterprise automation, deep tech,
and life sciences.
And in those areas, especially life sciences,nowadays, there's there's a lot more interest.
But but those areas, I think they'reunderfunded.
There's not a lot of growth stage managers thatcan underwrite them.
Yeah.
(42:02):
And we look
for the areas where that's the case, actually.
Mhmm.
So we we we do co invest with managers, and andwe have, like, a formula for doing it.
And we never ever wanna invest, co investwithin that with those that are not our
underlying managers.
So it's only funds.
Sure.
And then we never compete with them, obviously,in market.
Mhmm.
(42:22):
You know?
We will always do later stage situations wherethey they want us around.
Yeah.
No.
That's really helpful.
And you mentioned some of the sectors thatyou're interested in and focus on.
It's great to see some of them are also part ofyour background as well, you know, with your
industry experience.
So I see that often, and it's pretty typical.
But what are some what are some areas orindustries that you, see that's emerging, that
(42:47):
you that you wanna get into eventually at somepoint?
That probably is just not there like quantumand and fusion and all that's just to the you
know, right now, it's just been a lot of smokeand mirrors, but, anything else.
Yeah.
I mean, obviously, we balance, you know, aneconomic narrative with actual investable, you
know, asset.
Like, we
Yeah.
You know?
And there's certain things that are just notinvestable for us
(43:09):
at the moment.
Like, I wouldn't invest in quantum hardware.
Yeah.
It doesn't make sense for us to do that.
Where we see a lot of opportunities is reallyat the inner what we call sort of the next era,
which is transforming nature, which is at theintersection of, you know, computational
sciences and the physical and biological realm.
(43:29):
Mhmm.
I think it's more pronounced in the biologicalrealm where, you know, we're starting to see
the ability for these very large, you know,sort of multiomic data sets and for this
interface between, like, under under like,measuring biological and being able to then,
you know, test the biological and then be ableto develop things in c two and in silico at the
(43:50):
same time.
So I we actually think that precision medicine,precision therapeutics Mhmm.
And biology generally is the next engineeringplatform.
Yeah.
You know, you have the ability to sequence, theability to then now synthesize, and and then
over time, the ability to really, with highprecision, edit the genome or transform cells
(44:11):
for therapeutic use is gonna be so it's just soimportant in such a big market.
Yeah.
So we we're looking at a lot of things there.
And we have managed to focus in those areaseverywhere globally.
Yeah.
So we we care a lot about about that.
We we also care a lot about we've always careda lot about DevOps and infrastructure.
Generally speaking, you know, the nature ofsoftware development is continually changing
(44:34):
and requires, you know, a significant amount oftooling and infrastructure for, you know,
enterprises to compete.
I think that market is always gonna evolvedepend regardless of the foundational the
shift.
Mhmm.
So we look at that.
And then within within deep tech, generally, welook at different areas, everything from, you
(44:54):
know, new new like, advanced manufacturingtechniques.
We're seeing a lot in defense defensetechnologies, which which we care about as
well.
We look at some in cybersecurity.
We've we've seen aerospace has been an areawhere we focus.
We haven't done anything yet in terms ofsatellite technologies, but we've done other
(45:18):
things which with, like, aerial drone systemsand things like that.
So so those are the areas which which we tendto look at.
Mhmm.
The the main kind of, like, elephant in theroom, obviously, is sustainability, which is,
you know, the energy transition and, you know,sort of net carbon goals.
I think that is an area where we're stilltrying to understand where the big
(45:41):
opportunities are that are venture backable.
Yeah.
And that is sort of its own category versusbeing subsumed in a lot of what the venture,
you know, funds are doing anyways, which isMhmm.
Of these areas.
So Yeah.
We've said we've been climate funds for themoment.
Mhmm.
Because we don't we don't necessarily see whatwhere the big opportunities are, but, but
that's something we'll look at over time as asthat space evolves.
(46:01):
Sure.
On the manufacturing side, like the advancedmanufacturing side, can you double click on
that for maybe one minute in terms of what youthink maybe the materials are or the processes
are that you think are advanced?
Yeah.
I think well, I think the nature of ofmaterials, you know, will evolve.
Yeah.
We've seen some companies that we actually wejust came back from Germany, we see some
(46:22):
companies that are investigating new materials,computationally that have certain properties,
properties that can be used for all kinds of,you know, target end use cases from, like,
carbon capture
Yeah.
To sustainability, etcetera.
So I think, like, the nature of material Mhmm.
And the chemistry involved, I think, is gonnais gonna shift.
(46:44):
Yeah.
We've also seen companies that, create youknow, because I think the nature of, like, the
products that are being created in thematerials, are being serviced are servicing
very complex areas.
Mhmm.
So the nature of the materials are verycomplex, like in aerospace and things like
that.
And so you were seeing sort of just in timemethodologies for, you know, manufacturing very
(47:06):
complex goods at low volume Mhmm.
Near to the source where they're needed.
And so we think with all the reshoring that'shappening and and sort of, like, the move,
like, the deglobalization, I think that's gonnabe an area that's that's interesting.
Mhmm.
And then a lot of the, like, sort ofmanufacturing generally and, you know, that
area really hasn't, like, digital beendigitally transformed sufficiently.
(47:28):
Mhmm.
We're seeing, like, different softwaresolutions for connecting with controllers that,
you know, on on the plans and
Yeah.
Visibility
and optimization.
I think that's still, early days.
Sure.
Yeah, I've seen some interesting syntheticbiology applications where like people are
creating, you know, furniture and materialsusing like fungi and seaweed and stuff like So
(47:50):
but it's cool because it is a new advancedmanufacturing technique, and then it's also
kinda slowly getting into sustainability aswell.
So I was kinda thinking about that as youmentioned that.
But
Yeah.
We we've seen a lot in, know, justbiomanufacturing.
And, you know, there there there'scomplexities, obviously, not only in the
discovery side, but in sort of skip and yieldwhen it comes to, like, you know,
(48:13):
biofabrication and thinking about these things,fermentation, and and also quality control,
downstream quality control.
So we have stayed away from, like, pure play,you know, solutions like consumer solutions.
Mhmm.
You know, we like more platform level
Yeah.
Structure, like picks and shovels Mhmm.
(48:36):
Are our go to when it comes to these things.
Although I think amazing consumer companieswill be created, but we just
They'll be built on top of the platform prettymuch.
Don't we don't like to take consumer risk.
Yeah.
We say our brand development risk or go tomarket risk.
We like to take technical risk.
Yeah.
Absolutely.
Well, it's good.
Really good methodology.
Really appreciate the time.
(48:56):
And I think this is really, you know,insightful.
And we did portfolio construction live on thecall.
So appreciate you walking through a couple ofthose scenarios and have a good rest of the
week.
You know, this is really great, really goodconcept for the community.
So appreciate it.
And, hope to catch up in person soon.
Amazing, Joel.
Well, thanks for having me on the show.
Yeah.
Take care.
Bye.
For sure.
(00:00):
I mean, think I'll give you a little bit of ourmethodology, as you think will answer that
question.
And when we tackle when we looked at this spaceinitially, we said that we felt that it's first
of it's a very opaque market.
Yeah.
Welcome to The Investor, a podcast where I,Joel Palafinkel, your host, dives deep into the
(00:20):
minds of the world's most influentialinstitutional investors.
In each episode, we sit down with an investorto to hear about their journeys and how global
markets are driving capital allocation.
So join us on this journey as we explore theseinsights.
Okay, so we're live here.
(00:41):
I got a new guest, Albert Azut.
He's at Level VC.
So what I think which is really interestingabout their team and their platform is they're
using data and data science essentially to lookat funds and have a data driven approach in
their asset allocation strategy.
So that's why I thought I'd bring them on, youknow, because my community of fund managers and
(01:06):
LPs, we can just share notes live in real timewith the community and, and hopefully come back
with some interesting strategies across theboard.
But before we dive in, I want to introduceAlbert, he's the partner of the firm, and one
of the co founders as well.
So Albert, maybe you could just start welcometo the show, you know, excited about our new
(01:27):
friendship.
And maybe you can just start by just goingaround with your background and, and your early
career and how you got into fund to fundinvesting.
It sounds good.
Sounds good.
Well, thank you for having me.
Excited to be here.
Yeah.
So my my background, I'm I'm a softwareengineer.
So I guess, first and foremost, I went toschool in Boston.
And then, when I graduated, I went to work, onWall Street for about two and a half years as a
(01:49):
infrastructure engineer.
Mhmm.
You know, essentially doing hardcoreprogramming day to day.
And then I I left.
I was living in New York, I left to you know, Iwanted to start companies.
And so in New York, over the course of mycareer, I started a few companies in different
areas, all technical, some in in ad tech.
And then the last company I did was was focusedon machine learning operations.
(02:14):
So we were quite early in in the machinelearning movement, when it sort of, helping
enterprises move from raw data to machinelearning at scale.
And we were, you know, an early adopter of a ofa project called Apache Spark, and we
contributed to it.
And, essentially, what we did was we builtpipelining systems, for enterprises, which then
became the category ML Ops.
(02:35):
Mhmm.
And so we built that company, in New York.
I had a, you know, really strong team, and weended up selling the company to, Verizon in
2015 via via an AOL.
Mhmm.
And and that was an amazing process, difficultbut amazing process.
And as part of that transaction, you know, I Ihad the opportunity to to leave New York and
(02:55):
and go to California.
Mhmm.
So in in 2016, I picked up and I ended upmoving to to Palo Alto, California
Mhmm.
Where, you know, I worked for Verizon, forabout a year and a half.
In in Palo Alto, they have a, you know, anapply a research group applied research group
where, you know, the team works on, you know,very large scale machine learning problems
(03:17):
Mhmm.
Tackling very you know, sort of variousbusiness initiatives, you know, within AOL and
Verizon.
Everything from, you know, understandingwhether two devices are the same user to, you
know, predicting, you know, will somebodyconvert on a piece of advertising?
And all those are very, very hard problems.
So during that that time, I I was able toreally get a lot of experience in, you know,
(03:39):
scaling very large Mhmm.
Machine learning systems.
And then and then I started to get prettyheavily into deep learning and what became sort
of modern machine learning, what we see today.
And, you know, at that point, I I actuallyinteracted with one of the investors in my
prior company, and he was his name is Bobby.
He was putting together a new firm, in venturecapital, which which I believe had a very
(04:03):
interesting strategy, still does, which wasinvesting multistage, with a focus on
enterprise software.
So it's a firm called Coda Capital in SanFrancisco.
Basically, you know, they would invest earlystage, growth stage, and also public.
Yeah.
And that was an interesting purview on thewhole landscape because as you're investing in
private, you're getting sort of perspective ofpublic positions.
(04:27):
And so I ended up joining the firm, you know,as a partner in in 2016, and I stayed there for
four years, you know, helping build the firm.
And I was focused on the private investingside.
So all of the early stage investing and, youknow, my area of focus or at least, you know,
where I where I had the most appeal was in, youknow, deeper tech Mhmm.
(04:49):
Anywhere it's like semiconductor all the way toinfrastructure.
And then we did some application level thingsas well.
But that was really my my focus and and sort ofwhere I where I like to to spend time.
And then I did do a little bit more lifesciences, had more life sciences exposure
during the end of of that.
So I was there four years, and then duringduring COVID, ended up, you know, coming to
Miami, where this is where I'm originally from.
(05:11):
And Yeah.
We we we just were really happy here, and weended up just ended up staying.
That's Sure.
That's the short end of the story.
And in that, I had the opportunity to, youknow, transition out of CODA, which was an
amazing experience, but, you know, think aboutwhat I wanted to build.
And so that's how I ended up, you know, atbuilding level ventures.
Yeah, I see that often, I see a lot of buildersthat became VCs, including myself that are
(05:36):
starting to kind of build some data drivensoftware that you know, what happened with me
essentially was I became an I was in tech,worked in venture and still doing still working
in venture, but also starting to slowly build acouple products to help my workflows and then
just sharing it with the community and seeingif they they think it's useful.
If it's not, you know, maybe just iterating offof that.
(05:58):
And one of my favorite people is Yo, hey, Idon't know if you watch Yo, hey, on on Twitter,
he is he's at untapped ventures, and he buildslike a bunch of chat GTP, chat GPT algos like
in in real time on on Twitter.
So he he built like really cool, like, techparsers and and all that stuff.
(06:19):
So I think it's really cool.
And his his following just recently got a hugesurge because, you know, the content that he's
sharing, it's definitely it's definitely ofinterest to the community now.
Right?
People are people are getting into promptengineering.
So there's a lot to unpack and nerd out on.
You know, one of the first things is, you know,your your early days working on Wall Street
(06:42):
because I live in New York still, I've beenhere, you know, kind of like the way that you
think about Miami.
That's my family in New York City, you know, sowe live here, we've stayed here for about a
decade.
Tell me about the ecosystem when you wereworking on Wall Street and kind of like how
FinTech was developing like FinTech and data,data companies, because it's way different now,
(07:02):
right?
It's very vibrant.
So I'm just curious to know and I can tell youabout like how it was when I was there.
I worked at FactSet, you know, building a lotof their enterprise FinTech.
So it was kind of slowly developing, but lovelove some insight in terms of, how it was when
you were there.
Yeah.
I mean, it's a long time ago, so I'm I'm datingmyself.
But, you know, and I was I was really involvedin in sort of, like, core infrastructure
(07:25):
porting.
You know, I think if anything, can say that thecertainly, the architecture, you know, like,
sort of your traditional monolithic softwareapp and how you deploy software Mhmm.
And the speed and velocity at which you deploy,you know, it's just it's a whole different
world now.
You know?
The process from by which you develop software,you check it, you deploy it into this sort of
(07:46):
very monolithic process.
You know?
It's so, you know, sort of highly sort ofbureaucratic and regulated.
Yeah.
That was sort of one major yeah.
One that's a huge change.
Obviously, that's a consequence that's justchanged.
The time curve has changed a lot.
And then, you know, when it came toquantitative methods, which we did I did a
little bit of that while I was there.
Mhmm.
And I was I was getting sort of very interestedin, you know, quantitative finance and and
(08:09):
thinking about time series analysis.
But those methods, you know, are sort of verydifferent, than what you see today.
I mean, a lot of the same primitives are there,but Mhmm.
The world has changed in terms of how you howyou sort tackle those kinds of problems at
large scale.
One big thing I would say in terms of techteam, back in the day, the business pretty much
had all the power.
They're like, hey, you need to build this bythis quarter.
(08:32):
And then it's really like a project.
It's like, hey, this needs to be done by q1,like all these back end scripts or, you know,
the the platforms that you were building.
I feel like now the engineers have a huge stakeat the table.
And there's also discipline of productmanagement.
I'm assuming that they didn't have, like, theproduct management discipline back then.
(08:52):
So it's a bunch of like so was it like thebusiness stakeholders?
It's like, look, know, you need to or salesthat's like, hey, you need to launch this by
two months because the customer wants it.
Yeah.
Yeah.
I mean, it was it was basically the CTO makingdecisions and Yeah.
Down.
You know, I guess it was more of an IT functionand less of a strategic technology function.
Yeah.
I think it's been the major shift Mhmm.
(09:13):
Which is software is sort of the business in inmany cases.
And and back then, it was just sort of asupporting infrastructure for many of the other
things that the business was doing.
Some of it was critical, mission criticalinfrastructure, but certainly sort of forefront
in terms of the business itself.
And I think that obviously has changed, youknow, drastically.
Yeah.
And with that, I think the the organizationalstructure around, you know, product management
(09:37):
and thinking about engineering process, butalso just developers themselves are you know,
they have the ability to choose various toolsand tooling and infrastructure.
And since we're on the cloud, there's so manythere's such a combinatorial explosion of
things you can do to get software deployed.
Yeah.
Power has certainly shifted, you know, over thelast, like, twenty years or so.
Yeah.
No.
Absolutely.
I mean, the engineers have a huge influence.
(10:00):
They're not just kind of taking orders, theyactually, I think it really has to be an
aligned strategy with design, product marketingengineering.
And I think sometimes like the roadmaps need tobe aligned too because I mean, the engineers I
work with, know, they they wanted to continuebuilding interesting infrastructure and you
(10:22):
know, make the platform faster where like, youknow, I was a product person, right?
So I was more excited about adding new dropdowns or like new, new graphs and charts and
analytics.
And, you know, trying to get I think part ofthe product managers discipline is like,
getting the engineers bought sometimes justhelping them understand.
In my experience, like I burned the mostbridges.
(10:43):
Well, I haven't personally on the record, butbut like, I think engineers in general get
upset, like if there's throwaway code, right?
If they build something, and then it actuallycame to of no use to the customer and then they
have to like rebuild it again.
So I think in my experience, it's been and Ilove to hear your input on this, but like the
modern approach has been essentially, you know,thinking about the bigger vision of the
(11:06):
company.
So if you are going to go on mobile, have theengineers know because they could probably
choose the right infrastructure or like theright, you know, code and tools to kind of do
that where if you know you're never never goingto be on mobile, then there's probably a
better, you know, infrastructure to use or or,developer methodology to deploy if you wanna
(11:28):
achieve that goal.
And I I I agree.
I think if if know the endpoint on sort of, youknow, how you're servicing customers and
whatever they're gonna be on, I think it helpsmake decisions
Yeah.
A priori before you start building.
But at the same time, you know, there areplatform shifts.
Right?
And and Yeah.
You know, something like, you know, like LLMsor foundational models, I think those things
require sort of a rethinking and a re you know,sort of complete paradigm shift and design
(11:52):
patterns of how you build stuff.
So so I think, like, you can you know, as aslong as the teams are agile and they can sort
of there's not a lot of technical debt built upin the system, and they're able to sort of
quickly abandon things and, and sort of moveinto infrastructure that's more efficient, you
know, and and sort of and that's why I thinkthe breaking down of the monolithic app was a
big thing because
Mhmm.
(12:12):
It pulls you to, you know, very be more agileYeah.
Through how you develop.
But, yeah, I think those are sort of bestpractices that Mhmm.
You've learned, like, this learning rate thathumanity has had is, you know, in terms
of Yeah.
Software more efficiently.
And we're still really at the beginning of thatas well.
Yeah, no, absolutely.
And then, you know, I think it would beinteresting for the audience to to hear your
(12:35):
process in terms of building one of your youknow, I know you launched like four companies,
but maybe you can tell me, maybe the first oneor maybe, like, one that you choose would be
better to share, but essentially how you cameup with the idea and then how you got it to
market and
Yeah.
And
how you kinda scaled it.
Yeah.
I guess maybe this the last one is isappropriate.
(12:56):
Mhmm.
Yeah.
You know, I think the I guess when you're veryearly in an industry, and you're thinking about
problems that the market not necessarily isthinking of, so you you really have to, like,
pace yourself and, you know, have a a go tomarket motion and a sort of a product market
fit that needs to evolve over time.
Yeah.
As we're sort of recognize the problem becauseI think one of the dangers is like having to
(13:17):
teach customers.
Mhmm.
Well, it sort of creates a lot more friction.
Yeah.
So I think one of the things we learned thereis is when it comes to product management is
really just being on time in terms of the thesort of the problem set that you're that you're
solving.
But I think in building in building the productand iterating on it, you know, we had some
ideas as to, you know, the core technology thatwe wanted to build.
(13:41):
We felt that, you know, when you're dealingwith very, very large datasets and you're
trying to do something like, for example, youknow, unsupervised clustering, it's very, very
difficult as that dataset grows, and you needto have a methodology that's distributed that,
you know, you can sort of solve the problemefficiently even even with this, I guess,
you're, like, halfway there.
(14:01):
Mhmm.
So that's sort of where we focused, you know,our our efforts.
And so we had a really good core technology,but then there was a gap with the actual
product and how do you how do you into aninfrastructure.
So that's that was kind of a learning process.
Like, how do you Mhmm.
You know, architect a product that has that,you know, fits within a particular customer
workflow.
Mhmm.
And how do you go about selling that?
(14:23):
You know, you know, what's the productmarketing to it?
So that was a big big you know, and sometimesyou have to rework the whole product because
you just, you know, you UX based, and you justdidn't have the right fit.
So that was a very long process and a longlearning that we had there.
I think the outcome was good, but we werecertainly early in in the market.
Yeah.
Times is is is a disservice, you know, becauseyou need to be able to survive as well.
(14:47):
So Mhmm.
That was a big learning for me.
And also just building large scale machinelearning systems, I think, is is completely
different modality than, you know, than doingsomething that's just sort of traditional
software.
Yeah.
I mean, I'm sure the cost has definitely gonedown essentially with with usage based pricing.
(15:07):
But can you tell us a little bit about just howinfrastructure has evolved?
Like in terms, you know, because right now,when I think of data platforms, I know Google
has like Google compute, and there's likedigital ocean.
But if you're trying to build kind of a largescale data system, what do you need to think
about if you're like a founder?
(15:28):
Yeah.
And I think there's, you know, I guess there'sa few pieces of it.
And then like Hadoop was the thing that peoplewere talking about for a long time, you know,
distributed systems, but that was like, I, Iwas kind of in the space, like back in 2014.
So a lot of those buzzwords, like big data andstuff, I'm sure, like, a lot of that has
changed.
Oh, yeah.
I don't even hear I don't hear it as muchanymore.
(15:48):
Yeah.
I mean, Hadoop was an interesting solution forthe for the period of time.
I I I think it's it's much more overhead thanthan what's needed to manage data.
We see that with sort of flat data lakes.
Yeah.
More you don't need that sort of overhead.
It's more understanding the protocols and andsort of having the ETL piece of it.
But, yeah, I think all that wholeinfrastructure has changed.
And and, like, Hadoop was really meant toserve, like, the original distributed computing
(16:12):
framework, which is like a Macrius.
Yeah.
Came out of Google.
And, you know, that was the beginnings ofthinking about distributed computing in large
scale systems.
And I I really think there's a lot of the sameprimitives ended up being in in sort of some of
the deep learning techniques for for trainingthat we see today, obviously, with has its own
nuances.
So I I think that that certainly evolved.
(16:34):
But if if you're if you're a team building,infrastructure, data infrastructure, I think
you have to think about the, you know, thenature of the data.
Right?
Unstructured, structured, and and sort of howthat's stored and, you know, what are the
what's the schema for storage and what are thesystems for storage.
And then really, the data pipelining piece, Ithink, is so critical.
Mhmm.
A lot of the value I think you get besides, thedownstream modeling, which I think, you know,
(16:59):
you it's it's nuanced, but a lot of the valueyou get is really building data infrastructure
that is scalable, that's able to manage andintegrate datasets.
You know, for example, one big problem that wehave, as you can imagine, is is entity
resolution at level, right, which is, like, youhave many different datasets that are coming
in.
You have different entities in the dataset,like companies, people, etcetera, and it's very
(17:23):
noisy.
Right?
So you you're not you're not always surethere's ambiguity, but it's also very noisy.
So you have to be able to resolve a transactionor an entity or a company into the same thing
Mhmm.
With high, you know, high integrity.
And that is that's sort of a data engineeringproblem.
Right?
Sure.
At our scale.
So those are the kinds of things that you needto do very well and prepare the data in a way
(17:45):
that then could be used for modeling veryefficiently.
So I really think, like, a lot of the gruntwork actually, I think a lot of the grunt work
moving moving forward will also be very datainfrastructure heavy versus modeling work,
which I think will be sort of out of the boxreally strong.
So, anyways, I think that's, like, the keything is setting up
the
right the right pattern, the rightinfrastructure, the right visibility
(18:07):
observability in your data infrastructure.
And then from there, you know, I think is isreally, having I always feel like it's
important to have an intuition on what you'retrying to model.
Mhmm.
You know, you can you can apply, like,traditional out of the box stuff and say this
is you know, here's a target variable.
Let me try
to Yeah.
Solve it.
But I think you need to have an intuition onwhat the nature of the data that's being
created is, how it's being created What'sobservable about it, what's not observable
(18:32):
about it, and then model it based on thatintuition.
And then from there, you can choose models thatmight suit the purpose.
Okay.
And so I think that's something I've kind oflearned over time.
And not everybody has an intuition for howthings how they generate.
It's not easy.
But, you know, we've always that's the that'skind of like the modality and the the paradigm
that we've always operated under.
And then let's talk about kind of the new powerparadigm shift with, you know, auto GPT and
(18:57):
chat GPT.
Like, what what's kinda like the infrastructurethat needs to change?
I guess, are we are they still using datalakes?
And then and then I guess what's the humanimplication?
You know, how is this gonna impact maybe ourindustry?
You know?
I mean, can we can we ask prompts to look atcertain parameters on fund performance and then
(19:18):
have that pull up?
And is that kinda what you guys are thinkingabout with your, you know, your kinda data
driven approach and your data sciencebackground?
Well, I I think it's, it's certainly a aplatform shift, a significant platform shift.
I think the way we design applications Mhmm.
Moving forward is gonna change drastically.
If you just look at some of the primitivesaround you know, some of the things we're we're
(19:40):
talking about just today, like promptengineering.
Mhmm.
You know, it's like, that's a term that, youknow, really is, you know, just completely new.
Yeah.
Human or few shot learning or things like thatwhere you really have to think about, you know,
how do you build a system leveraging thesecapabilities in a way that provides value.
I think the way you build applications is nowgonna be completely different.
(20:00):
And you can also assume that asymptotically,you know, the systems will get a lot more a lot
more powerful.
The context window will get larger Mhmm.
And get cheaper.
You know, you're gonna be on a on a veryrapidly diminishing cost curve.
And so I think that's gonna make a lot oftraditional software development and
applications irrelevant Mhmm.
In many ways.
And I think you'll have a situation where,yeah, incumbents will integrate a lot of this
(20:23):
into their core application, and then someincumbents will no longer be relevant because
of we retrieve information, you know, how wecreate content and media and all that will
completely completely change.
Yeah.
That being said, I think it's you have to sortof take it with a grain of salt as well.
I mean, there it's not always good to, like,use, you know, historical, you know, sort of
(20:46):
corpus to generate future context.
Like Mhmm.
You know, you're gonna get these things likehallucinations and, you know, you're gonna get
sort of data integrity issues.
And at the end of the day, you know, the corpuson which it's training on is the human corpus,
which needs to evolve.
Right?
So Mhmm.
So we we, at least currently in its currentmanifestation, the way we're using it is more
(21:07):
to be able to embed entities into some sort oftopic space.
Mhmm.
So if I look at a company, I wanna understandvery, very low level, what is it doing?
Not not is it in health care, but is it doing,you know, DNA synthesis using enzymatic
approaches.
Right?
That's the kind of, like, understanding that wewanna have around companies and then founder
(21:29):
expertise and also company expertise and andthen also, you know, manager expertise, fund
manager expertise.
So we use it more for understanding the spaceof topics in which Mhmm.
Businesses are operating in.
Yeah.
At a very fine grained resolution.
I think we could eventually build a system thatcan integrate into our workflow so we can have
(21:50):
textual questions and responses, know, based onour own system of data.
So you you'd wanna combine your, you know, sortof large foundation models with, let's say,
your own internal datasets and knowledge Mhmm.
That you can actually, you know, extractknowledge from your own system in a much more
seamless way.
And then maybe, like, tie threads betweenthings you wouldn't have otherwise I've seen.
(22:11):
But I think out of the box for what we'redoing, we don't we don't you know, we're not
sure that it's the right approach to be ableto, like, make decisions.
Yeah.
That that's our current, you know, thinkingright now.
Well and I guess you can use, you know, us bothbeing from Wall Street.
You know, there's live stream data that you canpull in from, like, exchanges.
Obviously, we don't have any exchanges, but wehave data feeds.
(22:33):
So if we were to kinda build something thatcould be valuable for LPs, what do you think
they would want to see on their dashboard?
Like, if they're if they're like someone justtrying to do research on, hey, which which is
the best one?
Because there's so many of them.
A lot of the standard performance metricsthat's that benchmark, that's pretty much table
(22:53):
stakes, right?
So whether, you know, you're looking for acertain level of net IRR or a certain level of
ROIC and even an industry, right?
There's a ton of it.
There's a ton of funds focusing on AI.
There's a bunch of funds focusing on healthcare.
They all have the performance.
They're all great.
Right?
So so what else could we pull?
And, you know, if someone were build someonewere to build some type of data platform, how
(23:17):
could you determine kinda what is good?
And I guess Yeah.
Yeah.
For sure.
I mean, I think maybe I'll give you a littlebit of our methodology.
I do think we'll we'll answer that question.
And when we tackle when we looked at this spaceinitially, we said that we felt that it's first
of it's a very opaque market.
Yeah.
And there's a lot that's unobserved in themarket.
Right?
Mhmm.
And I like that.
(23:37):
That's an an opportunity we like, right, whichis a difficult one.
And then there's probably three things that aretrue, about venture specifically just given the
underlying, like, mechanics returndistributions.
One is that, you know, manager, performance ishighly dispersed.
Mhmm.
Right?
So that's it's traditionally, like, that's thecase where you have a huge dispersion in in
performance.
(23:57):
So if you're not really in the top quartile ofa fund manager, you know, you might
underperform the whole market, even even otherasset classes, which we've seen.
Yeah.
The second is that sequential correlation is isrelatively low.
So just because you did well in one period oftime doesn't mean you'll do well in the next
period of time.
To some some extent, there is some correlation,but it's not it's not extremely strong.
(24:18):
And the third thing is that it's very timedependent, you know, as we know, vintage and
cycle and background market conditions reallyaffect returns.
You have a lot of portfolios today where, youknow, they look really good.
But, you know, over the next few quarters asthese sort of 400 unicorns that have not raised
since 2021, you know, what happens to thoseportfolios, right?
And so we always had the intuition that thetrack record was certainly not enough.
(24:44):
Sure.
And it's an it's an indicator, but it's notsort of a strong indicator.
And so when we started to look at the market,we realized that, you know, one of the things
that, you know, what are the stable things thatdrive, you know, durable performance in any
market like this that's that's has a lot ofuncertainty.
And what we realized, yeah, a lot of it isnetworks, right, is is understanding the
relationships between entities, the flow ofinformation between them.
(25:08):
And if if one, you know, individual influencesanother person's decision, that actually is a
very strong signal.
And so we've taken the approach of trying toreconstruct, the network of behavior and
activity using what is basically raw data thatdoesn't have any network structure.
And so that's what we do.
(25:30):
U we construct features that are focused on,network, network durability, recency, strength,
etcetera.
And we use those features to then makepredictions on on future performance calibrated
against historical past.
The benefit of doing that is it's it'sinvariant to time.
(25:50):
Mhmm.
You don't get a you don't get a lot of thesesort of false positive effects of of purely
performance based metrics.
Yeah.
You actually realize the underlying driver isthe asymmetric access that some have in the
market, and we think that that's a big driverof that network.
So to answer your question, understanding theco investor base, understanding the sourcing,
(26:11):
the, like, the the discrete sourcing pipelines,the founder networks that they have, the those
kinds of things, I think, are so much moreimportant than looking at just, performance on
its own Yeah.
Which is just a side effect of those things.
No.
Absolutely.
I mean, I think that's also probably directlybeen related to some, you know, fundraising as
(26:35):
well.
Right?
And people that have been able closeallocations.
Obviously, there's regulations in terms of whatyou're allowed to share publicly, but just
people that are creating a community publiclybecause there's education that they're putting
out there.
Like if you're like the guy Yohei, I'lldefinitely ping his his Twitter.
You know, he posts a lot of content on AI, andthat's just built an engaged community.
(26:59):
And, you know, especially on Twitter that, youknow, there's a lot there's a huge engaged
community in the VC space there.
So I think using some of those social networks,I think, could help.
And is that what you're referring to?
Like, maybe some of the the data that you seeon Twitter, and then maybe also on LinkedIn?
And I mean, TikTok is up and coming too.
There's a few, you know, a few fun managersthat are really, really using that.
(27:23):
So I don't know if that's if that any of those,like, social networks are helpful as well
beyond Yeah.
I think it's a signal.
I think it's a signal for yeah.
Certainly, a fund has a flywheel aroundcommunity Yeah.
They're creating a lot of content, and there'ssort of this flow of of talent right now, I
think it's a very strong thing.
Yeah.
Although we we focus less on virality and moreon network, which I think is actually a
(27:43):
distinction between the two.
Because you could probably get information veryrapidly, but doesn't necessarily mean you have
durable network relationships.
Yeah.
So we
balance both of those things, but what I what Iwhat we're referring to more is very strong,
for example, co investor networks that that arehighly durable, let's say, or or access to
founder networks that are durable.
Mhmm.
It's not sort of one shot kind of things whereit really is.
(28:05):
Yeah.
Yeah.
That's what I'm talking about.
Okay.
And a proxy for that could be social because ifyou do have a big presence, then wanna be
working with you.
So that's important as well.
Yeah.
And you could probably find some well, I guesssome of that data would have to be probably
volunteered to, to be provided as they'reapplying to, you know, your your your platform.
(28:29):
And then some of that stuff you could probably,you know, just get on the Internet, you know,
just through data datasets and stuff like that,I'm assuming.
Right?
Yeah.
Yeah.
We we've always taken the approach that, youknow, from an evolutionary perspective, there's
gonna be more data, not less.
And especially in the private markets whereit's been very opaque.
There's more accessibility and democratizationof information.
(28:49):
Mhmm.
And so then what matters is not so much thedata itself, but, you know, sort of how you
manipulate it and insights are.
And and actually think networking the data isis a big, import it's an important element,
right, which is to take disparate datasets thatare orthogonal and put them together in a way
that's very unique.
I think that, like, that sort of thinking, Ithink, is more important than just getting
(29:13):
access to, let's say, Crunchbase or PitchBook,which everybody can.
Yeah.
And I'm saying that I think, you know, yourunique data science strategy to your point,
right, taking those sets and stitching it tocreate a new type of, maybe your own
performance score.
Right?
I mean, that could be really, reallyinteresting too.
It's kind of a novel because there isn't beyondthe traditional metrics that we've seen for
(29:34):
years.
I think there could be a network score or likea network performance score based on what you
guys have been, developing.
That's, that's really interesting.
And, you know, on top of that, you know, on topof the network, what's your, you know, for the
for the emerging managers that are applyingthat fit your criteria, you know, what are you
(29:54):
looking for in in managers and and what aresome of the things that managers should think
about when they're applying to, you know, yourplatform as well?
Yeah.
I mean, I think this we have an opinion on onthings that we look for.
Mhmm.
I guess there's a few different areas.
You know, one obviously is fund size.
You know?
Mhmm.
We're very strong believers that venture doesnot scale very well and that smaller fund sizes
(30:19):
have have big potential to outperform justrelative to the fact that, you know, you you
can return multiples of the fund with one exit.
Yeah.
And so we do look for funds that are reasonablysized.
Mhmm.
And and that's important.
And it's also an area which is very hard to todiligence.
Right?
Because Mhmm.
Usually, they have unless they stay purposelysmall, typically, they don't have a lot of
(30:40):
track, you know, to to Mhmm.
On, and it's more opaque.
Yeah.
And it's also underfunded.
So I think, like, that's one thing we look foris the right portfolio construction strategy,
including funds.
The other thing we we really like, more andmore is just special you know, to some extent
specialization.
Yeah.
We we believe that technology is getting morecomplex, and the the nature of the problems
(31:03):
that we're gonna be solving are more are verytranslational.
Meaning,
like, core science, either physics or biology,or even AI.
So these are deeper areas and translating theminto a business.
Mhmm.
And that we believe requires, you know, deepdomain expertise and and more so the ability to
attract entrepreneurs that are building thesemore complicated businesses.
(31:25):
Obviously, they wanna work with those that havehad experience in that area or understand their
businesses.
So we actually think specialization is isimportant.
And, like, it could be, like, domainspecialization.
It could be something else.
Right?
Yeah.
For the market specialization.
But so we do look for specialization generally.
Mhmm.
Especially in environment like today where it'syou just can't invest and expect good results.
(31:46):
Yeah.
So that's another thing we look for.
I would say flywheel, it's just hard toquantify.
Mhmm.
But there's something about, you know, like,every success and every investment you make
should have nonlinear effects on nextinvestment.
Right?
So it's not like you're just making randombets.
You're there's a, you know, a community angleor Mhmm.
(32:07):
An expertise angle or some other angle that'sbeing built over time, which is which is gonna
give you more results and nonlinear outcomes.
So we we look for that.
It's hard it's hard to quantify what that is.
Mhmm.
And then the last thing is just people.
You you wanna have sort of very intellectuallyhonest people that understand, you know, where
(32:29):
conviction can fail Yeah.
And sort of where you need to have diversity ofof investing and where you have to think about,
you know, how you treat entrepreneurs and howyou treat your investors and and how you, you
know, sort of how active you are or just activeyou are.
Just there has to be a thoughtful approach towhy you're doing things.
Yeah.
Anyways, those are just some of the things thatthat we look for.
(32:49):
In addition, we have we use our our technologyvery deeply, in our diligence to, you know, us
understand different features of the managerand, you know, based on their prior investments
and network and all that.
Art and art and science, I would say.
Mhmm.
No.
That's that's awesome to be building and also,using the tools yourself.
(33:10):
When it come I'd like to just double down alittle more on the portfolio construction.
So, you said reasonable size.
So what what what would be reasonable and thenkind of hit the board boundary of,
unreasonable?
And then I'd love to kinda have you double,talk a little more about portfolio construction
and your thoughts on that.
(33:30):
Yeah.
I think it's it's there's not a rule in youknow, there's not, like, a rule of thumb around
around these things.
I I think you wanna, you know, and it alsodepends on ownership, you know, like, your
ownership targets.
You know, I think there's really, at least inthe seed, there's really two, I think, two
kinds of investors or two two stable points ofinvestors.
One is sort of your positive sum game investorswhere, you know, they can fit into a round.
(33:51):
You know, they maybe they'll put quartermillion or half a million, and that's you're
not you're not sort of pushing anybody out, butyou could get on really, really strong cap
tables and that gives you a real return.
And then you have those that that are that aremore competitive and zero sum game where they
need to, you know, they need to deploy3,000,000 in a seat, and that that means that
others cannot.
And so those are the kinds of the two types.
Mhmm.
I actually think they require differentportfolio construction strategies for sure.
(34:14):
Mhmm.
I think for us, like, we the smaller, thebetter, to be honest.
Like, we we think, like, $5,025,000,000 is areally good fund if you have three companies.
Mhmm.
Where where you have, like, a relatively goodcost basis for ownership.
You you could manage a hundred and $50,000,000fund with with a good amount of, let's say,
fifteen percent first ownership.
(34:35):
Mhmm.
And let's say, a thirty thirty companies withwith with some follow on reserves Yeah.
Which I think is another really good state.
We call it a core position versus a singlemarket.
So I think those are, like, the two median sizefor us.
Like, our median fund size is a hundred millionin our portfolio.
We have 20 funds.
It's a hundred million.
Median.
I think that's the right that's the right sizeif you care about, you know, carry.
(34:58):
Right?
If you care about incentives, generallyspeaking.
So Yeah.
We we've done a lot of statistical analysisanalysis on this.
I mean, just on the the distribution of returnsand venture, and how you you can sample, you
know, a set of companies from a portfolio likeyou really wanna be in that in that size range,
I think.
Mhmm.
And, can you share a couple examples of a goodportfolio construction for a positive sum game
(35:23):
strategy versus a competitive zero sum gamestrategy?
Yeah.
I we we like so if you have, let's say, a50,000,000 or 25 to 50,000,000, that's like a
nano.
But let's say you have a $50,000,000 fund.
Yeah.
Let's do 50.
Let's do maybe like 50 for both, maybe to keepit simple.
Right?
Or 50.
50.
Yeah.
So 50 for a positive sum game, I guess.
(35:44):
How many companies?
How much reserve?
How much ownership?
Yeah.
You know, we we care about reserves to somedegree.
Mhmm.
I I don't I wouldn't we're not, like, sort ofoverexposed on reserves.
We actually have more companies.
Yeah.
But we we think that if you're sampling 30 to40 I say sampling, obviously, you're investing.
But if you're investing in 30 to four 30 to 40companies in a portfolio with, let's say, two
(36:07):
to 5% ownership, you know, then if you havesort of a very if you have a few sort of
nonlinear outcomes, I think that returnsmultiples Mhmm.
The portfolio.
Yeah.
And if you especially if you're recycling.
So, that's that's what we think.
So I I guess what I'm saying is, 50,000,000overly concentrated portfolio, let's say you
have 10 names Mhmm.
Makes us quite nervous.
(36:29):
Just because it's it's just you don't know.
This is a very uncertain environment.
Yeah.
It's it's, like, structurally uncertain.
It's not like we don't know.
It means that we we don't know what we don'tknow.
Yeah.
And so you there's no way to a priori know whatwill be successful, you know, no matter what.
And so Mhmm.
You have diversity in position sizing anddiversity in number of positions in your
(36:52):
portfolio, in order to be able to get thesesort of nonlinear outcomes.
You know?
You could get lucky, but that's not a strategy.
Yeah.
So that's what we think on, like, the$50,000,000, you know, positive sum game.
And then, you know, like, some of the firms wementioned before we started the the call, I
mean, that's that's what they do.
Mhmm.
And what what about the deployment schedule?
Are they looking to deploy most of the capitalin, the first three to four years and then and
(37:17):
then try to think about their next fund?
Or is it are they actually still deployingthroughout the whole the fund?
No.
They they typically have a three yearinvestment period.
Mhmm.
Yeah.
Especially for the smaller funds too.
They wanna get
yeah.
And you wanna get time diversity.
Right?
Because Mhmm.
You you know, a lot of, like, for example,like, you know, we know that LLMs are like a
(37:37):
platform shift.
We don't you know, I of course, we don't knowasymptotically what what the market structure
will look like in terms of value.
Mhmm.
But you wanna be able to take a few bets,whether in the infrastructure side or tooling
side or even, you know, sort of verticalizedapplications.
Yeah.
You really you know, you wanna be able to dothat over time, and you couldn't have done that
last year, and now maybe this a lot more.
So time diversity allows you to get exposure todifferent themes, thematic areas.
(38:02):
Mhmm.
And, also, you don't you're not sort of stuckon one vintage from a valuation perspective.
So I think you wanna have diversity.
Okay.
A lot of the great managers, I think, over thelast three years when things were very loose,
were the ones that paced properly.
You know, they were very, very thoughtful invaluations, and what they paid and how they
(38:23):
paced.
And so a lot of to be honest, a lot of ourmanagers haven't been deployed that much.
Yeah.
Our fund, you know, 10% has been deployed.
Mhmm.
You know, we're a fund of funds.
So Sure.
Which means that the managers aren't callingthat fast.
Yeah.
Which is why we like we like to see.
Mhmm.
So, yeah, that's that's what I would say aboutthat.
Okay.
And then the competitive who's here some game,how to I'd be curious to see yeah.
(38:47):
Yeah.
I'd be curious to see how that differs from I'massuming just a number of companies, I guess.
Or?
Yeah.
I think it's a it's probably a it's a largerfund size.
Think Yep.
You wanna be able to take larger stakes, youknow, in the company.
Okay.
So this will probably be maybe a 75,000,000 ifwe did the example here.
Yeah.
I'll say a hundred 75 to a hundred and likethat.
And there there you wanna have, you know,you're gonna have larger checks.
(39:09):
Right?
And you need to have some reserves.
I mean, a founder's gonna wanna make sure youhave reserves.
Right?
Mhmm.
They're signaling and other things.
And so it just requires a bigger fund.
And then you're competing with with you know,you need to have a team, I think.
Right?
And management fees matter, and you need tohave, a team of people around you to support
the companies because the founders, you know,that's you know, it's really good founders
(39:30):
understand who are the right investors.
Right?
Mhmm.
So so I think you need to have a biggerstructure in order to be able to do that.
Yeah.
And they're like, we never sort of vary from 20to 35 companies in a portfolio.
I think anything less is way over concentrated.
Sure.
It's it's very it's just too risky.
So yeah.
So that's what I would say.
And plus sometimes there's there's good extent,like, seat extensions where you can take more
(39:52):
ownership, and so you wanna have reservesthere.
And then Yeah.
Pay.
So I think there's 30, you know, 30 companies,let's say, twenty five, thirty companies,
hundred and 200, hundred and 20 five, you know,million dollar fund size Mhmm.
With, let's say, 50% maybe reserves
Yeah.
Is, like, what you typically see.
And think that could that could be a very goodstrategy if if you have conviction that they
(40:14):
can lead rounds.
Okay.
Yeah.
They that they can actually take the lead.
That's good.
That's good insight.
And then have you guys ever thought about doingsome direct investments along with your
strategy?
Because obviously, that's one of the biggestperks with some of the emerging managers and
just strategic funds.
(40:34):
You know, they they always leave some coinvestment opportunities.
So, you know, I guess, is that something youconsidered?
And then I've seen a lot of funds now slowlyhave, you know, kind of an LP, you know, on the
other, we'll we'll address this question first,but I've seen some funds also have kind of an
LP well.
They're like, you know, they mainly invest indirect companies, but they have a small carve
(40:57):
out to invest in some emerging managers too,which I think is great for for access to deal
flow and just networks that they normally don'thave.
Yeah.
Yeah.
So we we do a co invest, actually.
Mhmm.
We do a lot of that.
Cool.
It's always been our strategy to to support theunderlying portfolio companies later stage.
So, typically, that that happens with a seriesb or series c Mhmm.
(41:20):
Where the the size of the rounds are largeenough where we can, you know, come in
Yeah.
And and piggyback either on the relationship orthe pro rata of the manager.
And we share economics with our managers veryopenly.
So we do that all the time.
Mhmm.
And sometimes there's a series a extensionwhere we we we, like, we think the company is
really strong and just needs more time.
Yeah.
We're happy to, like, support that situation aswell.
(41:42):
Mhmm.
There's, like, asymmetry of information.
So, yeah, we we and, you know, where we focusis really an enterprise automation, deep tech,
and life sciences.
And in those areas, especially life sciences,nowadays, there's there's a lot more interest.
But but those areas, I think they'reunderfunded.
There's not a lot of growth stage managers thatcan underwrite them.
Yeah.
(42:02):
And we look
for the areas where that's the case, actually.
Mhmm.
So we we we do co invest with managers, and andwe have, like, a formula for doing it.
And we never ever wanna invest, co investwithin that with those that are not our
underlying managers.
So it's only funds.
Sure.
And then we never compete with them, obviously,in market.
Mhmm.
(42:22):
You know?
We will always do later stage situations wherethey they want us around.
Yeah.
No.
That's really helpful.
And you mentioned some of the sectors thatyou're interested in and focus on.
It's great to see some of them are also part ofyour background as well, you know, with your
industry experience.
So I see that often, and it's pretty typical.
But what are some what are some areas orindustries that you, see that's emerging, that
(42:47):
you that you wanna get into eventually at somepoint?
That probably is just not there like quantumand and fusion and all that's just to the you
know, right now, it's just been a lot of smokeand mirrors, but, anything else.
Yeah.
I mean, obviously, we balance, you know, aneconomic narrative with actual investable, you
know, asset.
Like, we
Yeah.
You know?
And there's certain things that are just notinvestable for us
(43:09):
at the moment.
Like, I wouldn't invest in quantum hardware.
Yeah.
It doesn't make sense for us to do that.
Where we see a lot of opportunities is reallyat the inner what we call sort of the next era,
which is transforming nature, which is at theintersection of, you know, computational
sciences and the physical and biological realm.
(43:29):
Mhmm.
I think it's more pronounced in the biologicalrealm where, you know, we're starting to see
the ability for these very large, you know,sort of multiomic data sets and for this
interface between, like, under under like,measuring biological and being able to then,
you know, test the biological and then be ableto develop things in c two and in silico at the
(43:50):
same time.
So I we actually think that precision medicine,precision therapeutics Mhmm.
And biology generally is the next engineeringplatform.
Yeah.
You know, you have the ability to sequence, theability to then now synthesize, and and then
over time, the ability to really, with highprecision, edit the genome or transform cells
(44:11):
for therapeutic use is gonna be so it's just soimportant in such a big market.
Yeah.
So we we're looking at a lot of things there.
And we have managed to focus in those areaseverywhere globally.
Yeah.
So we we care a lot about about that.
We we also care a lot about we've always careda lot about DevOps and infrastructure.
Generally speaking, you know, the nature ofsoftware development is continually changing
(44:34):
and requires, you know, a significant amount oftooling and infrastructure for, you know,
enterprises to compete.
I think that market is always gonna evolvedepend regardless of the foundational the
shift.
Mhmm.
So we look at that.
And then within within deep tech, generally, welook at different areas, everything from, you
(44:54):
know, new new like, advanced manufacturingtechniques.
We're seeing a lot in defense defensetechnologies, which which we care about as
well.
We look at some in cybersecurity.
We've we've seen aerospace has been an areawhere we focus.
We haven't done anything yet in terms ofsatellite technologies, but we've done other
(45:18):
things which with, like, aerial drone systemsand things like that.
So so those are the areas which which we tendto look at.
Mhmm.
The the main kind of, like, elephant in theroom, obviously, is sustainability, which is,
you know, the energy transition and, you know,sort of net carbon goals.
I think that is an area where we're stilltrying to understand where the big
(45:41):
opportunities are that are venture backable.
Yeah.
And that is sort of its own category versusbeing subsumed in a lot of what the venture,
you know, funds are doing anyways, which isMhmm.
Of these areas.
So Yeah.
We've said we've been climate funds for themoment.
Mhmm.
Because we don't we don't necessarily see whatwhere the big opportunities are, but, but
that's something we'll look at over time as asthat space evolves.
(46:01):
Sure.
On the manufacturing side, like the advancedmanufacturing side, can you double click on
that for maybe one minute in terms of what youthink maybe the materials are or the processes
are that you think are advanced?
Yeah.
I think well, I think the nature of ofmaterials, you know, will evolve.
Yeah.
We've seen some companies that we actually wejust came back from Germany, we see some
(46:22):
companies that are investigating new materials,computationally that have certain properties,
properties that can be used for all kinds of,you know, target end use cases from, like,
carbon capture
Yeah.
To sustainability, etcetera.
So I think, like, the nature of material Mhmm.
And the chemistry involved, I think, is gonnais gonna shift.
(46:44):
Yeah.
We've also seen companies that, create youknow, because I think the nature of, like, the
products that are being created in thematerials, are being serviced are servicing
very complex areas.
Mhmm.
So the nature of the materials are verycomplex, like in aerospace and things like
that.
And so you were seeing sort of just in timemethodologies for, you know, manufacturing very
(47:06):
complex goods at low volume Mhmm.
Near to the source where they're needed.
And so we think with all the reshoring that'shappening and and sort of, like, the move,
like, the deglobalization, I think that's gonnabe an area that's that's interesting.
Mhmm.
And then a lot of the, like, sort ofmanufacturing generally and, you know, that
area really hasn't, like, digital beendigitally transformed sufficiently.
(47:28):
Mhmm.
We're seeing, like, different softwaresolutions for connecting with controllers that,
you know, on on the plans and
Yeah.
Visibility
and optimization.
I think that's still, early days.
Sure.
Yeah, I've seen some interesting syntheticbiology applications where like people are
creating, you know, furniture and materialsusing like fungi and seaweed and stuff like So
(47:50):
but it's cool because it is a new advancedmanufacturing technique, and then it's also
kinda slowly getting into sustainability aswell.
So I was kinda thinking about that as youmentioned that.
But
Yeah.
We we've seen a lot in, know, justbiomanufacturing.
And, you know, there there there'scomplexities, obviously, not only in the
discovery side, but in sort of skip and yieldwhen it comes to, like, you know,
(48:13):
biofabrication and thinking about these things,fermentation, and and also quality control,
downstream quality control.
So we have stayed away from, like, pure play,you know, solutions like consumer solutions.
Mhmm.
You know, we like more platform level
Yeah.
Structure, like picks and shovels Mhmm.
(48:36):
Are our go to when it comes to these things.
Although I think amazing consumer companieswill be created, but we just
They'll be built on top of the platform prettymuch.
Don't we don't like to take consumer risk.
Yeah.
We say our brand development risk or go tomarket risk.
We like to take technical risk.
Yeah.
Absolutely.
Well, it's good.
Really good methodology.
Really appreciate the time.
(48:56):
And I think this is really, you know,insightful.
And we did portfolio construction live on thecall.
So appreciate you walking through a couple ofthose scenarios and have a good rest of the
week.
You know, this is really great, really goodconcept for the community.
So appreciate it.
And, hope to catch up in person soon.
Amazing, Joel.
Well, thanks for having me on the show.
Yeah.
Take care.
Bye.
Popular Podcasts
Las Culturistas with Matt Rogers and Bowen Yang
Ding dong! Join your culture consultants, Matt Rogers and Bowen Yang, on an unforgettable journey into the beating heart of CULTURE. Alongside sizzling special guests, they GET INTO the hottest pop-culture moments of the day and the formative cultural experiences that turned them into Culturistas. Produced by the Big Money Players Network and iHeartRadio.
40s and Free Agents: NFL Draft Season
Daniel Jeremiah of Move the Sticks and Gregg Rosenthal of NFL Daily join forces to break down every team's needs this offseason.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.