# About Zo Khalapyan

I've been building software since I was 10. I studied at UCLA, Harvard, and Stanford, spent years at Meta, and was early on Messenger.

After leaving Meta, I founded a high-frequency trading company. We raised $5 million and handled more than $1 billion in trading volume. But we couldn't find product-market fit.

While trying to bring some cash back into the company, I started exploring how to sell our codebase. That led me to task extraction: turning real software and engineering work into tasks that AI agents can learn from and be evaluated on.

What I began to understand was that the code itself was only part of the value. Making it useful for training—defining the task, preparing the environment, and establishing whether an agent succeeded—could create a more valuable offering.

I started this site to share what I'm learning. It's for founders exploring what their codebases might be worth, engineers building tasks for AI labs, and anyone curious about how post-training works in practice.

My longer-term belief is that post-training will become commonplace, with companies training agents on their own workflows. Task extraction could be a stepping stone toward that.

Tasks are an essential part of this process, but the practical work of creating them is still surprisingly opaque. I'd like to help make it easier to understand.


[LinkedIn](https://www.linkedin.com/in/zkhalapyan/) · [X / Twitter](https://x.com/0xZoZoZo)