Agentic RAG Advisor for Health & Culture Tourism
A RAG assistant that answers questions about health and culture tourism.
- Python
- LangGraph
- vLLM
- Qdrant
- Redis
- RabbitMQ
- FastAPI
- Next.js
- Docker
Approach
An agent that classifies intent, retrieves, checks its own evidence and only then answers.
- Intent classification routes each query before retrieval.
- Hybrid dense + sparse retrieval with Reciprocal Rank Fusion (RRF).
- Retrieval grading with re-query.
- Generation on vLLM, streamed to the client over SSE.
- Model side: RAG-aware SFT data and QLoRA fine-tuning, evaluated with RAGAS.
Architecture
01Queryuser
02Intentclassification
03Hybrid retrievaldense + sparse
04RRFReciprocal Rank Fusion
05Gradepass → 06 · fail → re-query
06GeneratevLLM
07StreamSSE → client
retrieval grading with re-query