Engineering notes
Insights
Writing about the problems we actually work on, including the ones where the honest answer is "do not build this". No filler, no keyword pieces — if we cannot write it with authority, we do not publish it.
- AI document processing architecture
LLM invoice extraction vs. template OCR: cost, accuracy, and where each breaks
Template OCR is better than a language model on the vendors it was configured for. The argument for LLM extraction is entirely about the long tail — and about which failure mode you can detect.
Read it - architecture billing state machines
Designing a membership state machine: freezes, grace periods and pro-rata billing
Most gym software models membership status as a handful of boolean columns. Here is why that fails, and what an explicit transition table buys you at the front desk.
Read it
Have a problem shaped like one of these?
We answer this kind of question on calls too, and the first one is free. We reply within one business day.
30 minutes · We reply within one business day.