2026
Arogya AI
FastAPI · LangChain · ChromaDB · Gemini · TypeScript
A medical RAG engine that knows when not to answer.
2 severity-routed collectionsStructured JSON enforced
Arogya AI is a stateless REST workflow over a medical knowledge base: when the initial context lacks symptom detail, the agent generates dynamic follow-up questions instead of guessing.
Retrieval is split into two ChromaDB collections — optimistic for general advice, pessimistic for diagnostic territory — and an automated severity assessment decides which one a query is allowed to touch. Multi-stage prompt guardrails plus Pydantic validation flag clinical red flags, block diagnostic hallucination, and force structured JSON output.
Key decisions
- Severity-based routing between general-advice and diagnostic vector collections
- Dynamic follow-up question generation when symptom context is thin
- Multi-stage LLM guardrails + Pydantic validation for clinical red flags
- Stateless REST design — every request carries its own context