Writing
Notes from a decade of building ML for chemical plants and factories. Modelling, software architecture, and lately how to make coding agents behave like engineers.
Posts
- Sep 2026
The best container runtime for a Mac, benchmarked
Five macOS container runtimes benchmarked on an M2 Pro across builds, filesystems, memory, networking, x86 emulation, and real Airflow and InfluxDB workloads. Method, medians with ranges, interactive charts, and every mistake logged.
- Jun 2026
Agents follow the rules you actually write down
How I run coding agents in a production monorepo: AGENTS.md rules distilled from failures, a library of skills, and MCP tools instead of longer prompts.
- Apr 2026
Parse, don’t validate, especially in a data-science repo
Why type discipline pays off faster in ML codebases than anywhere else: typed boundaries, no blanket suppressions, cheap refactors, and why coding agents make it pay twice.
- Feb 2026
When the neural network meets the mass balance
How hybrid first-principles and machine-learning models earn the right to steer real industrial processes.