Building AIsystems thattalk.

Four principles that guide every system I build, from voice agents to production web apps.

Ship first, optimise second.

Get it working. Get it deployed. Then make it fast and cheap. Premature optimisation is the enemy of progress. Evaluation-driven refinement is the ally.

Simplicity is a feature.

The best architecture is the one your team can understand at 2am during an incident. Complexity should earn its place, not accumulate by default.

Evaluate before you build.

Measure twice, cut once. I build evaluation harnesses before writing production code: state machines, retrieval benchmarks, latency profiles. Data beats intuition.

Production is the only proof.

Demos are easy. Serving real users at scale, with observability, cost controls and failover, is where engineering actually happens.

Where I've Worked

Results that speak for themselves.