Python
FastAPI
LangChainAn advanced Retrieval-Augmented Generation (RAG) system for multi-specialty hospitals implementing 7 sophisticated retrieval techniques. It features domain-specific knowledge retrieval across 8+ medical specialties with safety guardrails and patient-facing query restrictions.
Unique Features
7 advanced retrieval strategies: Query Decomposition, Contextual Compression, Self-Query Retrieval, Ensemble Retrieval, RAG-Fusion with RRF, FLARE, Adaptive Retrieval
Multi-domain medical knowledge bases (Cardiology, Pediatrics, Neurology, Dermatology, Endocrinology, etc.)
Automatic domain detection from query keywords without manual specification
Safety guardrails blocking harmful/threatening queries with patient mode restrictions
Dynamic retrieval tuning based on query complexity (k=2 to k=6)
~40% token reduction for simple queries with adaptive k-values
Tech Stack
BackendFastAPI, Python 3.10+, LangChain
AI/MLOllama Mistral (local LLM), HuggingFace sentence-transformers (all-MiniLM-L6-v2)
Vector StoreChromaDB
TechniquesRAG-Fusion, FLARE, Query Decomposition, Contextual Compression