Mokshith Voodarla, CEO Sieve: The Future of AI

all right I’m really excited to welcome MTH Vdara co-founder and CEO of Civ

where he’s leading the charge in building video AI tools for developers uh Moi has a really cool background

having worked in Ai and machine learning at companies like scale Ford Motor Company and Nvidia um he’s also a

graduate of UC Berkeley’s College of Engineering um and I think just as important to share Beyond his

professional achievements which are many um he’s also a fan of orange chicken Trace Le cake and the Boston Celtics so

I’m really thrilled to have him today with us to share his insights hopefully on you know Bill Russell LeBron James um

versus whoever his favorite team is and also talk about Ai and Innovation so so welcome M uh thanks for having me

Scott well um you know one thing that I I’ve thinking a lot about as uh AI

becomes more and more at the Forefront is like what the future of it looks like

in the the context of the world so like I grew up watching Superman cartoons with Brainiac and then data from Star

Trek um one of my favorite movies is her with you know um there’s David and

Prometheus like what’s your what’s your kind of world look like is it like robots is it people being supplemented

how do you see it I don’t know I I mean I think all of

those things are cool robots and and and you know augmented AI systems that you

know you can plug into and ask questions to when whenever you want but I don’t know one of the things I’m super

interested in and excited about is just kind of new ways to experience fun

things what I mean by that is kind of like you know what does it feel like to travel to a new place that you’ve never

been immediately um kind of virtually but but kind of really feel like you’re

there um and what does it look like to get models to maybe imagine those places

for you in this like very very realistic way or just new forms of entertainment

things like that that that are just much more immersive that I think like I don’t

know some some new generative models and things like that could unlock I think those are those are super interesting ideas to explore with um but obviously

robots are cool I’m someone that could do my dishes and and clean my room

that’s great but um yeah I I I’m more interested in some of like just the

cooler experience like humans can have um on top of the

productivity the productivity games okay so so I you know I don’t know if you’ve read the book or seen the movie Ready

Player one but I okay so so like the idea of going into the world you grew up

in in the past like seeing what your high school was like from a yeah I mean

you know sometimes I wonder like you know back in the 90s or the 80s like you

know um what did it feel like the excitement of calling up all your friends um to go

hang out and did that make hanging out more special um than than today when you just

kind of text and and you meet up whenever I don’t know um that’s that’s something I’ve always wondered oh well I

mean I could tell you it was exciting but also like you know if your friend didn’t answer because he was outside and he missed the call it was like your

right was ruined so it’s a it’s a really interesting one um so tell tell me a

little bit about Steve um you guys are past YC did you apply going in knowing

what you wanted to do or did you show up with like a kind of general idea and change it yeah so um you know my my

co-founder and I we we we started the company um straight out of school actually and um we were interested in a

set of ideas when when we kind of Applied actually like the idea we were most interested in was this kind of

concept of how do you kind of test and evaluate computer vision models better so the idea is like hey if you’re a

self-driving car company um I don’t know how well does your model perform when it’s raining versus Sunny versus windy

versus with a lot of glare um and we felt like I mean with some of the jobs that we done at different different

places we felt like most people didn’t really kind of care about evals enough

um and so that was the initial problem we were exploring we we got into y with

that idea we explored that idea for like probably half of the half of the batch and then kind of realized you know um a

lot of these tools that that that you know ml Engineers need you know there were like 20 other problems they were

trying to solve and this one problem kind of solved in in its own way was not

necessarily the right way to kind of go approach engineers and companies versus I don’t know bigger ecosystem maybe some

open source tool or whatever it was and so at some point maybe during the batch

we you know people were telling us about video um and the idea we had was around

kind of hey you know people have large video data sets this seems like an even harder problem than just you know

typical image data sets whatever how do you help people maybe find the most interesting moments in their data set to

go train their models um and I mean even that problem didn’t exactly stick but

but at some point I think the thing that clicked to us was like Hey we’ve been so obsessed with how do we help machine

learning Engineers um what if instead we start thinking about the problems

regular software Engineers face when they try to go do stuff with machine

learning um and kind of that switch of of of persona is is probably what I

would say kind of kickstarted towards what we do today kind of more focused on

developers broadly not just machine learning practitioners um and to focus on video and so so that was that was

kind of the story I mean my my co-founder and I we we got into a lot of the stuff doing robotics in high school

um and I mean yeah kind of got into computer vision that way and and then got into toola and things like that so

so so like were you guys I’m trying to remember what the program is called I lived in New Hampshire and like high schools near there they had those

programs where you could like create Bots that would try to kill each other were you that kind of stuff or it wasn’t

it wasn’t killing each other but it was first robotics um okay and you know

first robotics their their whole motto is the complete opposite it’s like this thing about gracious professionalism is

their term that they have and um I don’t know different games they have every year and there was this small there’s

like there’s always like some sort of a small computer vision component for for the games like anything from hey like

you know get the robot to like detect this green piece of tape that’s somewhere on the field and like drive

towards it and place it like a gear on it or something like that and so small tasks like that um um Obby my co-founder

is probably like he’s actually the first one who even taught me how to use open CV which is like this popular kind of open like just computer vision Library

um which is what we use for like a lot of these very simple tasks like tracking red balls green pieces of tape um things

like that super cool um when you when you guys started transitioning to like

thinking about software Engineers versus ml Engineers were there like like specific immediate obvious challenges

where you said okay we have our ICP our Persona kind of nailed what’s the first

thing we want to do like what’s the first feature product whatever I mean honestly it wasn’t super well thought

out when we were just trying things I I think like um the the first thing we did

honestly was just we put up a landing page and advertised a lot of different like video object detection use cases um

so everything from like the like videos of like pet tracking to like tracking

like angles in like a baseball game like there were like 15 different use cases I

think we just like thought of and put up and then it was like you could click you

could book a demo with us basically and then when when when you got on the call and you told us what you wanted to do uh

we’d be like okay cool and we’d get back to you in a couple of days with with basically the pipeline built for what

what you wanted to do um posted on our infrastructure and it was kind of this like it was like Consulting plus

infrastructure in a way for for a while and I think the things we sort of

learned there we were just trying to understand like why are these people even coming to us because we did get customers this way and like we solved

some problems and it was like you know the infrastructure is hard but it’s it’s not

as hard as like kind of this R&D step that I think you have to kind of do with

a lot of models where you know a lot of people would come to us with ideas where they’re like hey there’s this cool model

that does this thing like I want to use it um can I try using it and then when we gave them access to the model to use

they’re like oh like it doesn’t work in this case or this case and it’s like you know it it felt like we were guiding

them this like R&D process um and so it was like how do you make R&D easier for

engineers especially when there’s like a lot of different models out there they all have their kind of own tradeoffs you

know a lot of problems people solve involve multiple models you’re putting together in weird ways so I think like

those are the pain points we started like thinking about a lot more um as as to what we really wanted to Sol um kind

of through that process of Consulting almost totally makes sense um I I think

we had an investor early on who who would kind of give us a similar suggestion when we were thinking about new products or features it’s like throw

up a landing page you know put some ads towards I think that’s actually one of these things that like a lot of people

don’t talk about is just helping people figure out what they want is a big part of it and then you can kind of figure it

out iteratively yeah um so so like now that you guys have kind of locked in

this video AI for developers um you you one one thing I’ve noticed we’re I’m

personally in a slack channel that your company has or a customer of yours and it seems like once a week maybe maybe

more often than that you’ve come up with a new model you’re releasing something there’s a new feature and it just seems like a kind of almost like a Title Wave

of of stuff so I don’t know if your team is a thousand people or how you’re doing it but like yeah it’s changing quickly

and and how are you how do you guys think about like managing that whole process all these different tools resources

yeah yeah I mean I think it like naturally there’s like many many models

that come out every single day every week um and I think if we were serving

like a very very generic audience meaning just we called ourselves like an AI platform for

everything like I think it’s really hard to keep up um I think the thing that makes it easier is kind of understanding

what use cases our customers are most interested in and then figuring out the

relevant sort of building blocks almost that would enable our customers to build exactly what they’re trying to build and

I think that like kind of hones in a lot more as to what we should spend our time

even working on optimizing or like building pipelines around whatever it is

so I think like just having more focus is is super important and I think like a lot of our use cases are in in kind of

digital media video whether it’s social communication creation right um and so

there’s like this realm of use cases that um are most commonly used in our

product and I think like starting there is super important um and so focus is is

probably the best answer I have there what what about like internally uh you

guys are presumably also using AI to build your business like so you you’ve you’ve got this like Warehouse of AI

features and products that other companies are releasing or that you’re you’re you know being made open source you’re pulling from those what are you

guys doing internally to like be fast like move be Innovative yeah

um I think I mean a couple of things one thing that we found generally with any

kind of AI feature people build is that you know there’s there’s always going to

be like 10 20 different use cases but like the interesting pattern you notice when you kind of focus on on some Niche

is that there’s a lot of components that are super reusable across different use

cases and what I mean by that is like I don’t know one example I could give you that’s super tangible here is like um

dubbing um which is like this pipeline that’s on Civ where you can translate videos into different languages um a

component of that involves um you know tracking who’s speaking and then also

applying lip syncing to to kind of make it look like they spoke the other language the interesting thing is that

that same thing to track who’s speaking can also be used um for other pipelines

that are kind of focused more on content repurposing and reformatting so the idea of like hey I have a podcast recorded in

a landscape kind of camera can I just focus and autoedit the videos to focus

on who’s who’s actively speaking uh which is very useful in like you know podcast long form kind of content and so

I think like there’s a lot of efficiency gains that can be made if you really

treat problems as hey like what are the reusable pieces um that maybe are

applicable to other problems that can go be solved as well and so we find ourselves kind of of reusing stuff which

I think is like one of the big big things that allows us to move move more quickly and I think it’s also like kind

of something we’ve tried to bake in to the product experience we’re building and so all these things that we make

available out of the box use the same kind of product experience that our customers can also use to build custom

applications in the product as well so I think that that philosophy is super important very cool um I was just

thinking about the dubbing example I woke up one morning and I think one of

our Engineers had made like 10 different videos of a welcome to zeit okay intro

video and he had like put it into like polish Spanish and and like I was showing to my kids they just thought it

was such a like a delightful fun experience yeah super super cool um what what about things that like you’re

hopeful to see like where you’re kind of wa you you know maybe you’re not going to develop it but you’re waiting for it

are there things where you’re kind of excited to see that haven’t yet come to fruition I

think I mean one Trend I I I want to see kind of across the board is just

everything we already have just faster and cheaper um I think there’s just a lot of you know you I mean text to

speech is a great example of this where there are great Texas speech models that

can sound like me that you know run quickly but they cost too much um today

for example uh what would it feel like for you know I don’t know being able to

apply textto speech dubbing to every video on the internet um how can you get kind of costs so low that you know you

can enable those sorts of things or I don’t know even like you know classifying a large library of videos

and actually looking at every single frame to to to kind of have some granular search um right now like I mean

people sample frames people might like have weird tricks to kind of reduce cost but I don’t know I think like just

generally the trend of faster cheaper but everything we have today um is one thing um I’m excited about another kind

of pattern I’m excited about is just like models that let you

control um kind of outputs a bit more easily what I mean by that is like I mean again the example of Texas speech

is like you know it’s one thing to clone your voice and and have the AI speak like you it’s another thing to you know

know maybe be able to control the specific accent or the specific emotion

or things like that really easily right um and so I think just generally like

and and you see this even with like some of the new text image models like I think um there’s this new model that

came out called like flux um which I think is integrated into Twitter now through grock um and that like generates

these like very realistic looking images um but I think they could be a lot more

work done around like control as to like hey I’m explaining to the model this

exact thing like how well can you follow my instructions um so I think control

and then just speed and cost um across the board are are are things I’m excited

about yeah I would I would have to agree especially with speed or or even like connectivity I I’ve noticed for example

that outside of the workday like particularly when I use open AI chat GPT

like 5 6 PM with my kids they’ll say hey Dad what’s the fastest land animal we

start talking and it’s like sorry can’t connect I see I just I think there’s still sort of like infrastructure pain

you all these sort of little things right um but what so you guys kind of went from all right we’re going to go

out there and figure out some cool um API video products you’ve got customers

now what are some of your like I don’t know like when when you look at it or look at or talk to some of your customers and you see what they’re doing

what are some of the examples where you’re like like whoa this is like I can’t believe they did this or this is totally different than what I expected

us to be doing yeah I mean I don’t know one of our customers that that that we work

with that I think is super interesting is is is a company that um you know they do a bunch of video gen stuff um but

they’re very focused on the music industry um and creating really nice

visuals for like music artists um and the interesting thing there is like it’s

not necessarily even about making these like effects and cool patterns look real

it’s about making them like react to the audio um and like change and warp um and

I think some of their work I mean they have like a a you know a consumer product that anyone can use to kind of

generate these visuals but um they also kind of have the studio component of

their product that um you know artists big artists go use and I think create

visuals for like their Coachella sets and and things like and and that’s super

cool to see because it’s like wow like I didn’t think any of any of the stuff we we we we helped people build would would

go play at like Coachella I’ve never been um and so that’s that’s one

interesting thing um have you thought about asking them to pay you in Coachella tickets

perhaps I mean I have not but but um I

should H that’s that’s a great use case any others you think are super cool

um I think I mean like OB I think the obvious ones are like I mean dubbing is is obviously super interesting right

just because like I mean it is reducing sort of um language barriers um across

the board and I think like there’s great use cases that we serve in education and

and in content creation broadly um that I think are really cool um another one that I think is like

pretty fun is uh we recently released um a really optimized version of Sam 2

which is this like video segmentation model um that you know you can point

give it a point and then it’ll track that point um or object across a video

super super accurately and there’s some super interesting like sports tracking use cases I think um where you know we

even kind of gave it points of like a ball and it even you know when it the

ball goes behind a person in like this soccer game and things like that um it still tracks it super super well and the

ball is like this really tiny thing on the screen um and so I think like some of the sports stuff is cool I’m I’m I’m

a basketball fan and so um I think some of the sports use cases are really cool um yeah none others off the top of

my head um those are those are some that stand out those sound awesome um you one

thing we talked about was was how you know you guys have an investor um he was

previously CEO of pars he’s at Matrix now Ayah and um you know thinking about

when when I was working with him at pars and we were building a developer tool I think one of the ways that our Engineers

developer team you know developer marketing support we all thought about making a great product was actually having incredible documentation um and I

think companies like Stripe have done the same thing where like people look to them how do you guys feel like um you

you attempt to or try to kind of stand out in presumably a competitive space is

it is it documentation is it features is it being a head innovating yeah I mean I think this is

something that like we’ve learned the hard way over time where like developers have a very very high kind of quality

bar when it comes to products that they use um and they

like can get very suspect about you know the smallest things that they may not have control about control over um in in

you know whatever tooling that they’re using um I think like one one I don’t I

don’t know just like a general kind of philosophy that I I mean I was talking to someone last night that that they

they kind of put put these words in where was kind of like how do you make a developer actually feel like the hero uh

when they use your product rather than you seeming like the

Savior um right because like and and and I think this comes from building a

product where the onboarding experience and like the things you make available out of the box like are super simple and

and super easy to get started but that not necessarily meaning that too much is

abstracted away from the developer where they should still have more kind of

control um even if not everyone wants it I think they should have the option to

kind of dive deeper understand the internals um of how things work um rather than those

kind of being black boxes and so I think like one of the big differentiators that we think about is like you know there’s

a lot of these like very pure sort of API products out there that focus on very specific use cases um but a lot of

the time the thing that you end up missing sort of using some of these products is that it’s just the only part

of the developer experience there is hey like make an API call get a response fact that’s all whereas you know we have

some of these out of thebox apps on Civ um that are really great Hooks and and ways we get developers into the platform

but really like a lot of the philosophy is like developers can understand what’s going on under the hood they could take

certain components of these application to kind of go build their own things um

and so it’s more of like a tool that you can work with rather than hey like you know response in response out um so I

don’t know like those are those are some of the ideas we we kind of think about a lot and I mean this is still a work in progress for us like we’re nowhere near

what I would say is like you know um a tool that you know millions and millions

of developers Rave about and and we want to get there and I think like this comes from focusing a lot on product quality

and then kind of these these kinds of I don’t know psychological things as as to you know how you even put yourself in

their shoes like and how you would feel um using using a product like this so

it’s a great answer I love it um let me ask you a couple questions about you know just kind of wrap up on on I guess

more on just leadership and and how you think about being CEO or or a leader or founder within your business um

obviously you’ve got a lot to do I think we talked about how you’ve got you know some marketing that you’re working on right now and you know you’re you’re

doing all sorts of interesting really cool things longterm and then short term how do how do you figure it out or do

you just work 24 hours a day are you like how do you make it work what’s what’s your approach to making it work

and like finding the things that are most high leverage most impactful um I don’t know I mean this is

also a thing I would say I’ve I’ve struggled with a lot just because I mean this is my my you know first job um

technically right and and and I think um the the things I’ve I’ve learned is

mostly just that you got to hire people you trust um

and you know like as long as you are setting kind of Direction and and really

good kind of you know be being critical about like even the smallest things when you just don’t like them I think people

kind of generally understand um what that means for what they’re doing and

then like if they you know they’re talented they also have their own opinions um like and if they’re just

better at that specific thing like you should kind of let them do that thing that doesn’t mean you shouldn’t like you

know um I don’t know like I think what was it the Paul Graham had the article

founder mode right um but yeah I don’t I don’t know I think that it’s kind of

like yeah I abstracting away the things that like make sense to abstract away

just cuz like you know you can’t focus on everything but then still like just doing the things that you’re good at

um I think for me it’s like I try to set aside some time every couple weeks to just like be like man like what did we

even do these last like three weeks like was it worth it like did like is is it

actually going where where where I want it to go um I think those kinds of gut

checks are super helpful um where where’s your place that you I mean like

you know you talked a little bit about being a Celtics fan like do you go to Celtics game to kind of like just relax

and let let wash over you where do you go do you go to the mountains like what’s your no no no I mean I actually

really like I mean two two places I really like going so I live in SF and um

I don’t know like my housemates uh one of them actually had this routine where every couple of weeks he’d go like on a

Friday or Saturday night to like a very specific wine bar and he would just go

read just read there uh the The Vibes there were very much just like you can read you could talk to people whatever

it is um and and I don’t know that specific place I really like and I’ve gone there and just written things um in

in a notebook before but then I also just really like kind of uh walking um

there’s this really nice grass area in the marina um that I just really like walking through um and just like

thinking working out I don’t know working out super critical right it’s like it seems like such a basic human

thing but like without it I think you go czy as a CEO yeah exactly so um do so so

like going to your back to your point about you know sort of diving deep into the business occasionally um when I was in Facebook uh you know most of the VPS

that worked for Mark Zuckerberg talked about this eye of Sauron uh where he would sort of like you know gaze out

into marketing or growth or whatever and go deep and really understand the details and then okay he’s coming out

and he he’s he’s going back I I think just like you said it’s it’s almost impossible to know everything all at

once yeah and so it sounds like the approach you have is actually uh pretty similar across the board for for

companies where you kind of find an opportunity and go deep on it yeah yeah I mean do you have any advice

here I I think like one of my favorite things to do is it’s either three or four weeks I mean so like the approach

that I take is like we have probably a bi-weekly company all hands I have like a monthly pirate metric SAS worksheet

that I upd update manually myself like it’s probably 100 lines of data that I

comb through and just sort of see how the the company’s doing we have our board meeting so it’s basically just

checkpoints of reflection and you know I I know we talked about having a board or or whether it’s your board of your two

co-founders or like me we have you know a couple of our investors having some form of accountability yeah to force you

to look at things more critically super helpful and to give you advice to look at your team because like you got to trust your team but you also need help

looking at them right right makes sense well um one final question uh you know

it let’s let’s look at the magic ball the future um so we talked a little bit about the type of world that you see you

know maybe it’s a sort of a Ready Player oneish like being able to zoom in and out of the worlds that you’d like to visit uh what what else do you see out

there what’s exciting like even in terms of like technology that is is coming

soon I think I mean one of the things that really got me into machine learning

broadly was was robotics right um I I mean first robotics was the thing that

got me into computer vision and I think like there’s a lot of just cool stuff

people are pushing the bounds on with with humanoids um I think that that that’s super interesting where I mean a

lot of the robotic that’s out there in the real world is very much manufacturing these like very specialist

systems um something that I I mean the whole kind of goal of humanoids is that

hey like you know you have these special systems today and you have these whole R&D teams that are like building these

very specialized systems or specific tasks you know with the humanoid if you

actually build you know a versatile enough machine um the idea is you could

go apply it to any task and so you don’t have to invest R&D into every single new

problem you kind of build the thing once and you can go apply it anywhere now

it’s hard to say like how far away this stuff is and I think a lot of the constraints are very much in um I don’t

know like actuators Motors and and kind of manufacturing them um the costs of

that and there’s only a few companies out there that can actually go do that at scale um I don’t know I just think

like the idea of like robots um and humanoid robots is really cool um I I

don’t know if you saw I think it was Nat fredman um he he he basically said I

would love to find a way to build a robot that picks up the leaves in my backyard quietly in the middle of the

night and he was like if anybody wants to do it I’ll give you 50k and like it’s like six months have gone by and he’s

now he’s got this tiny little robot that picks up three leaves at a time and I just think about stuff like that where

it’s I I really just don’t want to do it like it’s those kinds of things I’m excited about the future I saw a robot

the other day that like cleans my walls or does my dishes uh I I have to snowplow my driveway every now and again

it’s just I see all this stuff that is fun but brutal yeah yeah yeah so

robotics broadly is interesting um yeah I mean maybe maybe one day i’ I’d love to kind of get back into doing robotics

at some point so well one thing you could potentially do at some point to get away from the grind is go teach kids

how to be like what you were and and kind of bring the Next Generation into what you’re doing yeah yeah awesome man

well hey thanks for joining me today and I’ll see you soon awesome cool