MEASURED / FIELD NOTES FROM BUILDING AI SYSTEMS

AI systems,
with the work shown.

I lead engineering for AI systems, and I publish the measurements behind the decisions.

Head of Architecture & AI Platform at Tvavium.ai. Previously founded a healthcare AI company and led engineering teams at Atlassian, Expedia and Akamai. These are my own views.

I publish a field note when an experiment has a result: the setup, the numbers, and the decision it changed.

INSTRUMENT 05 / LIVE GPU

Did adding a medical LLM make search better?

Replay the verified run or run it live on a GPU, then browse every query, including the answers the model broke.

Open the medication-resolution lab →

Most benchmarks answer the vendor’s question.

Building AI systems means making decisions no leaderboard covers: which model, which policy, how much memory, what quietly breaks. Each note measures one of those on a frozen setup and shows the work, so the evidence makes the call.

From the notebook

All posts →