Congrats on the $50 million Series A fundraise. It's a huge opportunity for you guys. What I thought we could start with is diving in, for folks who aren't familiar, to the Chalk story and what you guys are trying to accomplish. Give us the TL;DR on who you are and what problems you're tackling.

Thank you. We are very grateful for the opportunity to be here and also to have this new round. Chalk is the data platform for inference. Everyone's talking, obviously, a lot about model training. People spend a huge amount of time and money training their models. The thing is, once you have a great model, it's something that you want to run again and again, maybe forever.

What's really interesting is that everyone knows more and more compute is being allocated to AI. I think what's a little more subtle is that compute is obviously being allocated to training, but compute allocated to inference is really exploding. What I mean by inference is when a user clicks on a button, when they load a page, can you make a decision in that moment? That is what we're really focused on.

There are some amazing solutions in the market today. Obviously Databricks and Snowflake are really focused on complex training jobs. If you have a huge amount of data and you want to do a big batch process, that's something you might do overnight. But what if you need to serve an answer really quickly when a user clicks? You can't kick off one of those jobs for billions of users in milliseconds.

So what people wind up doing is caching data. They'll put it in a feature store like Google Vertex or Amazon SageMaker, where you have this cache of pre-processed data. So it's not fresh. It's fast, but not fresh. For companies that care about fresh and complex, there's this awkward point in the middle where they wind up trying to build their own thing. That's what we're solving for. Instead of having to build your own thing, Chalk helps companies make complex decisions at inference time by providing a really, really low-latency solution to do some really incredible things that effectively bend the curve of what's possible.

Let's drill into that a little bit, because you mentioned Databricks and some of these established platforms that folks have been using. What is the core innovation or differentiation that you guys have built that enables you to be much higher quality at the inference level?

Historically, a lot of stuff has been done in batch. That means you pre-process a huge amount of data for millions or billions of customers all at the same time. You can imagine traditional money movement, something like ACH. It is a daily batch process. You queue up all those checks, and then once a night you run that job, you check for fraud, and you move the money. Same thing with something like product recommendations. Look at Netflix, which is famous for doing a really great job there. They spent a decade doing an overnight batch job where they would look at each user and predict what you might want to watch next.

But fast forward to today, and people want to move money quickly, instantly. Even think about debit cards, which are not that new. You need to make an immediate fraud decision. You don't have overnight to think about it. Also think about recommendations. Obviously Netflix is doing a great job today. They update the home screen based on what you just clicked on, the preview you just watched two seconds of. And think about one of our customers, Whatnot. They're the largest live streaming marketplace in the United States. They moved from a batch job of updating their home screen to an immediate job of updating it based on what you were just browsing and looking at, based on Chalk.

So we're really helping companies that want to make those instant decisions, where you pull really fresh data and you run the model in response to a user action. Just to set the context, what I mean when I say inference time is when a user clicks, when a user loads a page, in that moment, we can kick off a job and get data to your models to make a really fresh decision. What everyone was doing before that was these batch overnight processes, where you would process millions or billions of users all at the same time, instead of one user making a prediction, a decision, on demand.

You've described a very broad variety of use cases, pointing to customers in e-commerce, in payments and transactions, and in other swaths of use cases that need access to real-time data. One of the questions a lot of folks have is about their broader AI tech stack. There are a bunch of different types of models that might be best catered to their specific use case, and they're trying to leverage them in a number of different environments or in production. How should folks think about you? Is there a lot that you're trying to replace for them? Are you trying to be as accessible as possible and let them use whatever model they want? Where do you fit into that stack?

Two answers. We tend to work with really mature companies, including some really large venture-backed tech companies and some Fortune 500 companies, so the whole gamut. We don't want to go in and replace everything on day one. Typically we have a very focused starting point, which is, "Hey, I want to launch a new product. I want to change my user signup flow from overnight to instant. I want to get faster recommendations out to my users." We're doing this with ApartmentList, with Turo, with Doppel, with MoneyLion, with Mission Lane. So it's a really broad range of customers where we're helping them make faster decisions.

So one, we have a really focused starting point. But if you really ask about the capabilities of Chalk, Chalk is really a feature platform. Something that's differentiated and really exciting is that we help with real-time inference decisions, but part of what we do is let our users write simple, natural Python, because that's the language of data science. Everyone looks at that and says, "Well, Python is slow. If we're writing Python and we've got to resort to this batch job, how can we take Python, put it in production, and go really fast in the way that you're saying is possible?"

That's part of the magic of Chalk. We take Python and translate it directly into C++ and Rust, which are the fastest languages you can find. So data scientists and data engineering teams can write Python once, and we'll transpile that into a really fast, performant language like C++. You can literally run your Python at inference time in a really low-latency way. That's the scope of things that we're working on.

Walk me through your hypothesis for how companies are evolving. There's a specific aperture of folks that fit your ICP, if you will. It's not only folks who need real-time data that is highly engaging and interactive with the user, down to the milliseconds you're able to provide. It also sounds like it's generally larger, more sophisticated companies processing massive volumes of data that make this really valuable for them. You get a smaller and smaller aperture of folks. As you project into the future, do you think more companies will want to leverage real-time data in a way that makes you increasingly valuable to a broader universe of companies? What's your hypothesis for how the world will look over the next two, three, four-plus years?

I'll share a little bit of our background and how we got here. Our backgrounds are all in fintech. I have two amazing co-founders, Andy and Elliot. They met when Elliot was a freshman at Stanford and Andy was a senior in high school. He came to admit weekend and slept on Elliot's floor, and they have been best friends and working together ever since, which is totally silly, but true. After school, Andy went to Palantir. Elliot was one of the first 10 engineers at Affirm, and when he was there, they said, "Yo, Elliot, we want to take a lot of data to feed into our models to make these instant lending decisions at checkout. Buy now, pay later. Go build our data infrastructure." And Elliot did. Then they had a startup together called Haven, a neobank that they sold to Credit Karma. Both at Haven and at Credit Karma, they found they were building the same things. So it's a very tried and true way of starting a company: "I built this before, I have some real insight about this problem, and we think we can do something really good so that not every company needs to rebuild a solution if they want to make real-time decisions."

To your question, our backgrounds in fintech showed us that real time was not just optional. If you want to move money quickly, if you want to make an instant lending decision, if you want to verify a user, their identity, or anything else, that has to be instant. But I think the whole world is realizing that when decisions are faster, when you rely on fresher data to make predictions, the user experience gets so much better that users love your product more, they spend more time, you make more money, and everything works better.

So yes, our background informed our starting point, and we're lucky to have a lot of customers in the fintech, verification, fraud, and identity space. But today we're actually doing a really, really broad range of things, because we think these inference-time decisions matter for everybody. Databricks obviously serves a wide range of customers that are doing some really complex training jobs. We are aspiring to be a next-generation Databricks focused on inference, which really means that if you have data that you want to operationalize and get to your models for ML and for AI decisions, we are a really great solution for that kind of work.

You sit at a really interesting vantage point. One of the big questions folks have about everything going on right now is that there's a ton of hype and excitement around different models, the capabilities they're providing, and what AI can unlock for a bunch of different use cases. There are a bunch of large enterprises amassing huge budgets to figure this stuff out, get it into production, and really benefit from the advantages there. The biggest bottleneck folks end up focusing on is that it's still a really difficult thing to do: to take a really cutting-edge capability and get it into a business use case in a massively complex environment with huge volumes of data. From your vantage point, what are the biggest limitations? Are you finding that these big customers have all their data cleaned, understandable, and ready to go, and they just need an engine like yours to make it happen? Or are there a ton of other bottlenecks these folks are struggling with as well?

There is a lot of hype. There is a lot of excitement, and for good reason, because the possibilities with AI and with data are limitless. But you're going to need infrastructure to do it. That's why, as we approach the market, we were really lucky to find an amazing partner in Felicis to help us with the next part of our journey. They looked at Chalk, and we look at this problem as: listen, no matter what you want to do, infrastructure is going to be here. It's the nuts and bolts that you actually need to bring your dreams to life.

Obviously a lot of companies have a huge amount of data, and people have been spending a long time figuring out how to take their data and make it really impactful for their business. That's still a work in progress. But we actually think the hardest part of the problem is getting data to your models, for training them and for serving answers. We think that operationalizing of the data is the hard part of the problem, and that's why we're so focused on that piece. We think that's the difference between actually being able to make your dreams come true and just having an idea that you'd like to see.

Is there a category that you feel is really primed for this use case? You talked about how your background in fintech really set you up, and payments and transactions happen not only at high volume but also at rapid frequency and with really significant complexity. Fintech is a really great training ground for high-volume, complex, real-time data. Are there categories that you think are the next fintech, where you're seeing customers, in e-commerce or wherever, who are really primed and ready for this use case? And are there other categories where real-time data could be valuable, but the category writ large isn't ready for it yet and some things need to happen to make it possible? I'm curious what categories you're most excited about.

I was lucky to work at Google for a bunch of years, and at Google we really thought about speed as a feature. When search got faster, when things loaded faster, when the internet worked better, people would search more, ultimately click on more ads, and have a better experience. I think that thinking really applies to pretty much every business.

As I mentioned, we're lucky to help with a lot of lending and fraud decisions around money movement with companies like Mission Lane, MoneyLion, and Melio, a really wide range of companies there. But we're also increasingly doing stuff with identity, with Socure, with Persona, with different kinds of fraud applications, with Doppel, and with consumer marketplaces like ApartmentList, Turo, and Whatnot. We just signed up one of the largest logistics companies based in India, which is helping to route packages.

So it's really hard to pick. I think real time makes everything better, and data brings those things to life. The question is: "Okay, we have great ideas for how we can use AI and how we can build a model, but how the heck are we going to get all the data? Are we going to have to do an overnight job? Is it going to be expensive? Are we going to be serving stale answers? Are we going to ask you to wait a day to sign up, wait a day to verify? Or can we make it instant?" And if we make it instant, will everyone have a better experience, and will we make more money? We think the applications are pretty limitless. I just mentioned a bunch that are top of mind, but we're really excited to see what's next.

Final question for you, on how you think about the future opportunity for Chalk and the key milestones you're really excited about. When you and I jam again a year from now, what do you think you'll be most excited to brag about having accomplished or built out?

I'll give you two answers. One is that we really live and die by the customer stories that we enable. When we're able to help a customer with one use case, go to a second, and really help enable their business and help them grow, there's nothing more exciting and rewarding for us.

Of course, we're also really focused on our own business. Last year, we set out to triple our revenue, and we did 3.5x year over year. We're ready to do that again. So we're very focused on building a big business. We've got a lot of growing to do and a lot to live up to, with our customers that are really trusting us and our investors that have given us this really big opportunity to make a really big long-term company.