Hi, I'm Prashant.

I started out as a backend engineer. APIs, services, databases, the usual. A few years in, I started working with machine learning on the data side, mostly because someone on the team needed help and I was curious enough to keep going after I'd helped. That turned into doing more ML, then more, until at some point I wasn't really a backend engineer anymore. I'm based in Calgary.

Most of my last few years has been LLMs. RAG, fine-tuning open-weights models for specific tasks, evals, and the agent stuff that's been showing up in real production this past year. The title I use now is AI Architect because most of what I do is figuring out how the pieces fit together rather than writing all the code, though I still write plenty of code.

In 2025 I finished a book, Machine Learning to Gen AI Agents. It's 18 chapters plus appendices and a glossary of 177 terms. The idea was one book that takes you from classical ML through to modern agents without skipping the boring parts in the middle, since the boring parts are where most projects actually go wrong. There's a print edition and a Kindle one. The audio edition is in production.

Outside of the book, I'm building a few proofs of concept. The main one is a privacy gateway. There's also a stock predictor and a multi-agent platform, both earlier. All of them are working POCs, not companies. The details, demos, and decks live on the POC page — that's where to go if you want to dig in.

One opinion I'll volunteer here, because it shapes everything else: I think most teams spend too little time on evals. Models are easy to swap. Eval suites that actually predict production behaviour are not. If you're building with LLMs and your eval setup feels embarrassing, you're in the same boat as most of the field, and it's the highest-leverage thing you can fix.

Outside of work I read a lot, mostly non-fiction. Calgary is a good city for that.

If you want to reach out — about the book, about Avarna, about something you're working on, about giving a talk somewhere, or just a question — my email is at the bottom of this page. I usually reply the same day.

A short journey, in landmarks.

Four cards. The agent adds new ones as new things ship.