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
- Jun 2026
Agents follow the rules you actually write down
The model is rarely the constraint, context is. What I have learned running coding agents on a production monorepo, from AGENTS.md rules to skills and MCP tools.
- Apr 2026
Parse, don’t validate, especially in a data-science repo
Parse data into typed structures at the boundary, fix type errors instead of suppressing them, and refactor without fear. Types turn out to help coding agents as much as humans.
- Feb 2026
When the neural network meets the mass balance
Neural networks extrapolate badly and first principles lack fidelity. Hybrid models cover each other’s weaknesses well enough to steer real processes.