The Threadmaze journal
Project writeups and field notes on building practical, privacy-first AI and data systems.
A first-person look at Overwatch, the in-house real-time alerting platform built at Jackson Health System: three use cases, ascending stakes, and the line that explains why most organizations never build one.
No-shows, missed calls, and lost leads in field-service businesses aren't five separate problems. They're one problem, no owned data store, wearing five different masks.
Computerization made everyone an accidental secretary. AI is making everyone an accidental manager, and the durable edge is knowing where mistakes come from in your business, not knowing the tools.
A cross-encoder reranker is the standard next step for a RAG pipeline. We evaluated one on a real retrieval product, it confidently ranked the wrong documents first, and the replacement was a scoring function you can read in one screen.
97% of health leaders believe AI is essential. Almost none feel ready to deploy it. The reason lives in fragmented EHR data, not budget or appetite, and having a FHIR endpoint doesn't mean that's solved.
The four-part RAG pipeline in plain language: why the compute is cheap and CPU-bound, why document preparation is the hard part, and why owning your data layer beats owning the model.
How I built a retrieval-augmented benefits assistant that answers plan questions with cited sources, refuses to guess, and never sends data to the cloud. The first node of a distributed private-healthcare-AI idea.