Notes from the
engineering trenches.

My wife asked me what the end goal is for my open source cancer research project. There isn't one. What the project became this summer, why the medical field is skeptical of AI driven research, and why I'm going to keep going.

A dozen people and about a hundred sensors when I joined. Thousands of sensors around the world today. The road between, as my commit history remembers it.

How we took an API from a few thousand requests a month into the billions — and why the hardest part wasn't throughput, it was the migration nobody noticed.

Shipping LLMs into real production systems — not demos. What it took to trust a model enough to put it on the hot path, and the guardrails that made it safe.

A side project that turned into 4,830 papers, a simulation engine, and a book. What building it taught me about evidence, humility, and shipping research in the open.

Every broker I reached for was wrong for the edge in a different way. So I started writing one that treats power loss as the normal case — durable, self-healing, one static binary.

Redis runs the world's caches on a single thread. What does a from-scratch, shared-nothing, thread-per-core cache that still honors the Redis contract actually do?