AI product and engineering, shipped behind evals.

A working surface, not a deck. Built with the same primitives the AI infra world ships on — retrieval, traces, telemetry, and the patience to put a real eval gate in front of them.

The shape of the work

Evals & quality. Reproducible eval suites, regression gates, and the discipline to ship behind them.
Retrieval & RAG. Vector search, chunking, recall metrics — the parts of LLM systems that quietly decide outcomes.
Observability. Tracing, telemetry, and the feedback loops that turn model behavior into product signal.
Product judgment. Translating fuzzy AI capability into roadmaps, runbooks, and risks a team can actually decide on.

Try it, don't read about it

Explore the archive — a deep essay archive with semantic search and article-aware answer surfaces.
Boot the immersive layer — a real-time 360 surface, not a screenshot.

Building something where evals and retrieval matter?

I'd rather sit with your roadmap for an hour than send a deck. If you're hiring for AI product, applied AI, evals, or platform work — let's talk. Reach me at alexwelcing@gmail.com, on LinkedIn, or GitHub. Full background on the about page.