What this site is
I work forward-deployed: I build AI systems inside other people’s businesses, next to the people who have to live with them. It’s a strange, great job. You see exactly where the demo ends and the product begins — usually about two weeks after everyone clapped.
This site is where I write that gap down.
Most of what’s published about LLM systems comes from two places: research labs, and people selling something. There isn’t much from the middle — the person wiring a model into a fifteen-year-old ticketing system, negotiating with a legal team about what the assistant is allowed to say, or explaining to a room why the eval suite matters more than the system prompt. That middle is where I spend my days, and it’s full of patterns worth sharing.
So expect practical notes, not takes:
- Evals — how to know your system works before the customer finds out it doesn’t.
- Retrieval — what actually moves the needle when “just add RAG” meets real documents.
- Agents — where they earn their complexity, and where a
forloop was fine. - The last mile — integrations, guardrails, observability, and demos that close rooms.
Client work is confidential, so I write in patterns: real problems and real numbers with the serial numbers filed off. If a post reads like it’s about your industry, that’s because the same failure modes show up everywhere.
If any of this is useful to you, email me — I answer.