Anand Kulkarni

In 2016, I was looking at early literature about not large language models, but the precursor technology known as recurrent neural networks. I remember reading this great paper by Andre Carpathy, who was at Stanford at the time. It was called The Unreasonable Effectiveness of Recurrent Neural Networks.

And he showed that with a pretty small program, you could train on a very large amount of data and get something produced that looked almost, but not quite like code. And I looked at this and I said, you know what, this is the future. As this technology improves, this is going to be the way that everyone ends up building software.

And he showed that with a pretty small program, you could train on a very large amount of data and get something produced that looked almost, but not quite like code. And I looked at this and I said, you know what, this is the future. As this technology improves, this is going to be the way that everyone ends up building software.

And he showed that with a pretty small program, you could train on a very large amount of data and get something produced that looked almost, but not quite like code. And I looked at this and I said, you know what, this is the future. As this technology improves, this is going to be the way that everyone ends up building software.

So we set out to build a company that would automate the process of software creation using AMP. We took the approach that we were going to have to build our own training sets and models from scratch. We set out to build essentially a data collection, the ability to capture information about how software was being written.

So we set out to build a company that would automate the process of software creation using AMP. We took the approach that we were going to have to build our own training sets and models from scratch. We set out to build essentially a data collection, the ability to capture information about how software was being written.

So we set out to build a company that would automate the process of software creation using AMP. We took the approach that we were going to have to build our own training sets and models from scratch. We set out to build essentially a data collection, the ability to capture information about how software was being written.

And the goal here was to understand how requirements, the things that we use to describe how we build software mapped into libraries so that we can understand the relationships between how people described code that they wanted to build. and then what libraries, technologies, and solutions they wanted to use.

And the goal here was to understand how requirements, the things that we use to describe how we build software mapped into libraries so that we can understand the relationships between how people described code that they wanted to build. and then what libraries, technologies, and solutions they wanted to use.

And the goal here was to understand how requirements, the things that we use to describe how we build software mapped into libraries so that we can understand the relationships between how people described code that they wanted to build. and then what libraries, technologies, and solutions they wanted to use.

It turns out the best way to get that kind of data is specifically by running your own development marketplace. And so we ran an open development marketplace for many years. It took us about five years to get all the data we needed. The deal we made was that if customers needed software built, they could come to us. We would match their needs up against development teams who could do the work.

It turns out the best way to get that kind of data is specifically by running your own development marketplace. And so we ran an open development marketplace for many years. It took us about five years to get all the data we needed. The deal we made was that if customers needed software built, they could come to us. We would match their needs up against development teams who could do the work.

It turns out the best way to get that kind of data is specifically by running your own development marketplace. And so we ran an open development marketplace for many years. It took us about five years to get all the data we needed. The deal we made was that if customers needed software built, they could come to us. We would match their needs up against development teams who could do the work.

Customers would keep the code that they built, but we would keep the metadata. The association between requirements and what libraries were being used to satisfy those requirements. And that let us build out this really unique, interesting data set.

Customers would keep the code that they built, but we would keep the metadata. The association between requirements and what libraries were being used to satisfy those requirements. And that let us build out this really unique, interesting data set.

Customers would keep the code that they built, but we would keep the metadata. The association between requirements and what libraries were being used to satisfy those requirements. And that let us build out this really unique, interesting data set.

And of course, that was the core of how we ended up building this classical machine learning system that would map requirements into code and code into requirements. Brownfield is the overwhelmingly biggest and harder problem to solve, and it's mostly neglected by industry. We realized that was really the core audience for this.

And of course, that was the core of how we ended up building this classical machine learning system that would map requirements into code and code into requirements. Brownfield is the overwhelmingly biggest and harder problem to solve, and it's mostly neglected by industry. We realized that was really the core audience for this.

And of course, that was the core of how we ended up building this classical machine learning system that would map requirements into code and code into requirements. Brownfield is the overwhelmingly biggest and harder problem to solve, and it's mostly neglected by industry. We realized that was really the core audience for this.

When we talk about the MVP, we're really talking about two generations of this company, right? Yeah. Era one was building to collect data and helping people build Greenfield software so that we could collect data that we needed. That era was the first five years of the business. We built the whole stack in Python directly with Django on the back, At the time, it was a view on the front.