Keeping the model away from the numbers
Why the AI in Navanto’s risk briefings picks the charts but never fills them in, and how deciding what a feature is allowed to get wrong shapes the architecture around it.
Read
Writing
Notes from building AI into a product people rely on.
Occasional write-ups of decisions I’ve had to make and the reasoning behind them. Mostly about where a language model earns its place in a system, and where it doesn’t.
All posts