The unglamorous parts of AI product engineering, done carefully.

I write about what it takes to build AI systems where wrong answers are expensive: looking at your data, determinism, logging and traceability, getting text out of PDFs.