Ras Patel · personal
HEALTHCARE · Clinical reasoningResearch

Zebra Diagnostics

Graph-native diagnostic decision support — Bayesian inference over a clinical knowledge graph.

ROLEFounder
YEAR2025 — now
STACKFastAPI · SQLModel · React · TanStack Query · Docker · Fly.io
Zebra Diagnostics architecture: patient records arrive over FHIR via qhint, are normalized to SNOMED CT by SNOAPI, and feed docdoc's patient graph; bray computes Bayesian posteriors over a SNOMED/LOINC knowledge graph and docdoc's Brief Studio shows the clinician brief, all orchestrated by ligature

Zebra Diagnostics is the company I'm building full-time. The mission is to save 200,000 lives in 8 years by making medical diagnosis less of a coin flip on the first visit.

The reasoning core is `bray` — a Bayesian inference microservice that takes patient evidence, looks up priors and likelihoods from a SNOMED CT– and LOINC-aligned knowledge graph, and produces patient-specific posterior probabilities over candidate clinical problems. Around it sits `docdoc`, the React Brief Studio that renders the clinician brief, evidence timeline, and CTA feed.

The system is designed for interoperability — clinical data flows in via FHIR (see `qhint`, the Redox/QHIN onramp), terminology is normalized through `SNOAPI`, and `ligature` orchestrates the Docker-based medical microservices behind a single control plane.