Asror Wali
I build AI products — and the websites and apps around them. I write about what it actually takes to make them work: evals, error analysis, and prompt engineering.
About Asror
I build things on top of LLMs — agents, chatbots, and the workflows around them — and I build the websites and apps they live inside. Most of my writing is about the unglamorous part: working out whether the thing actually works, and why it doesn't.
On the AI side that means evals you can trust, error analysis on real traces instead of vibes, prompts that hold up when users do something you didn't plan for, and knowing when the fix belongs in the prompt versus the pipeline around it.
On the product side it means the rest of the stack — schema, API, and interface — because an agent nobody can use is just a demo. Some of what I've built is public on GitHub and collected under Projects.
So the site splits four ways: Agent AI for putting this stuff in production, Full Stack for building the thing around it, Projects for what I've shipped, and Notes for everything too short to justify a post.
If you're shipping something in this space — or you want to argue about an eval setup — email me. I answer.