The hybrid deterministic-LLM pattern for healthcare AI
Why the most reliable healthcare AI systems use language models for interpretation and rule engines for computation, and how to architect this in practice.
·9 min read
Open sourceApache 2.0Healthcare AI
The healthcare AI community deserves better tools.
Open source tools and writing for engineers building AI in healthcare. Synthetic PHI generation, agent evaluation, and open data standards, alongside essays on the architecture that holds up in production.
Apache 2.0 licensed·No account needed·Built in public
Synthetic PHI Generator
1 / 5 recordsNAMESarah ChenNAME
DOB04 / 12 / 1987DATE
MRN847-293-105ID
DXICD-10: I25.110
CPT93306
INSAetna PPOGRP 4872
NPI1234567893
HIPAA Safe Harbor · Client-side onlyTry it live →
2Published essays
3Open source tools
Apache 2.0License
100%Client-side tools
Recent essaysAll essays →
Open toolsAll tools →
Healtheval
LiveNamed failure modes and evaluation prompts for healthcare AI agents. Catch hallucinations, misrouted prior auths, and fabricated codes before production.
Synthetic PHI Generator
LiveGenerate realistic fake patient data for healthcare AI dev and test environments. Fully client-side, no data leaves your browser.
Benefit-JSON
Coming soonAn open JSON schema for representing health plan benefits, copays, deductibles, coverage tiers, and prior auth rules, in one clean, machine-readable format. Not just a tool. A proposed open standard.
Subscribe
New writing, every two weeks.
A new essay every two weeks on healthcare AI architecture, RCM patterns, and the technical realities of building in this space. No marketing, no spam.
Or find it on GitHub.