Clinical Agentic RAG
LLM agents inside real clinical workflows.
In production
- Live
- Mobile app integration
LangGraphvLLMQdrantLangFuseAWS Lambda
The problem
Generic LLM chat is easy; an LLM system that operates inside a clinical workflow is a different discipline. For a digital-health product, nutrition guidance has to be personalized, grounded in clinical knowledge, auditable — and delivered through a mobile app, not a playground.
The system
Agentic RAG pipelines built around LLaMA, automating three production jobs:
- Nutritional guidance — retrieval-grounded answers over clinical nutrition knowledge.
- Meal plan generation — structured, personalized plans rather than free-text suggestions.
- Sentiment classification — routing and prioritizing user messages.
The agents are orchestrated with LangGraph, retrieve from Qdrant, and serve through vLLM. Every generation is traced in LangFuse — in a clinical context, observability is not optional. The whole system ships as AWS Lambda functions consumed directly by the mobile app.
What production taught me
Results
- Live in production — integrated into a mobile app via AWS Lambda
- Full generation tracing with LangFuse
- Part of the patented health-tech stack at Hibiscus Health