TriageTide
Explainable, Local-First Symptom Triage Prototype
Built a safety-oriented Streamlit prototype that converts structured symptom reports into explainable urgency guidance using deterministic local rules, explicit consent boundaries, and documented limitations.
System Architecture
Safety-oriented triage architecture derived from the TriageTide repository.
Structured symptom report
The guided form collects symptoms and boundaries before any optional storage or AI processing.
Emergency signs first
A versioned deterministic rule engine checks high-risk warning signs before narrower patterns.
Urgency decision
Rules select emergency, same-day assessment, or monitoring without claiming a diagnosis.
Reason + next action
Matched evidence, rationale, limitations, and concrete next steps are shown to the user.
Isolated integrations
Groq intent routing and Supabase storage remain opt-in and outside the core decision path.
Three explainable urgency levels from an inspectable local decision path.
Project Overview
TriageTide is an educational symptom-triage prototype for water- and food-borne illness scenarios. It prioritizes the safest next level of care - emergency, same-day assessment, or careful monitoring - without claiming to diagnose a disease.
The core decision path is a small, versioned, inspectable Python rule engine. Emergency warning signs are evaluated before narrower symptom-pattern rules, weak evidence remains weak, and every result explains the recognized signals and why the recommendation changed.
The primary guided assessment runs locally without external services. Optional Groq processing is limited to chat-intent classification, while optional Supabase storage requires separate explicit consent and is isolated from the medical decision path.
The repository includes 19 passing automated tests across application smoke paths, rule behavior, adversarial symptom combinations, negation, consent boundaries, password handling, and safe CSV export. It is a software prototype and is not clinically validated or regulator-approved.
Problem
Overlapping early symptoms can make users either dismiss serious warning signs or place too much trust in a confident disease prediction from incomplete information.
Solution
Designed a conservative, explainable triage workflow that assesses urgency before symptom patterns, keeps medical decisions local and deterministic, separates optional AI and storage consent, and communicates limitations prominently.