Product
Sehat Guftagu
Bilingual Urdu and English clinical assistant built on a multi-agent LangGraph backend.
Sehat Guftagu is a clinical web app for patients who are more comfortable in Urdu than English. It handles triage, symptom analysis, and report generation through voice or text in either language, and routes anything urgent to a doctor rather than answering it.
The reasoning is split across 5 specialized LangGraph agents rather than one prompt, so triage, symptom analysis, knowledge retrieval, report writing, and escalation are separately testable.
Clinical flow
- Patients interact in Urdu or English, by voice or text.
- Emergency triage detects red flags and escalates instead of continuing the conversation.
- RAG over a medical knowledge base grounds the answers.
- Consultations are written up as SOAP reports.
- A doctor portal lets clinicians review, approve, and prescribe.
Agent design
Five modular LangGraph agents handle the pipeline. Model output is parsed into structured objects with entity extraction rather than being handed back as free text, which is what makes storage, review, and the doctor portal possible.
Data
PostgreSQL through Prisma for records, Pinecone for the vector index behind retrieval.
Stack
- Backend
- FastAPILangGraphPrismaPostgreSQL
- Models
- Groq LLaMAPineconeElevenLabs
- Frontend
- Next.js 16React 19TypeScriptTailwind CSS 4