Recommend health check packages based on user information.

🩺 Health Check Recommendation (HCR)

A multi-agent RAG system for personalized health check package recommendation, built on the AgentOS framework and powered by DeepSeek V3.

Features

• Hybrid Retrieval: BM25 + dense vector retrieval with Reciprocal Rank Fusion
• Cross-Encoder Reranking: BAAI/bge-reranker-v2-m3 for precision improvement
• Medical Synonym Expansion: Ontology-aware query expansion (30+ synonym groups)
• Query Decomposition: Automatic splitting into symptom / history / demographic / risk sub-queries
• Multi-Agent Architecture: 4 specialized agents coordinated by an Orchestrator with reflection loop
• Dual-Mode: Switch between multi-agent pipeline and legacy single-agent mode
• Evaluation Framework: Retrieval, recommendation, and safety metrics with ablation study support
• Streamlit Web UI: 4 pages — Recommend, Chatbot, Hospitals, Report

🗂️ Folder Structure

HCR-by-AgentOS/
├── config/
│   └── settings.py                  # Configuration (hybrid, rerank, synonym settings)
├── data/
│   ├── health_check_data.csv        # Health check package database
│   ├── symptoms.pdf                 # Medical symptoms reference
│   ├── medical_synonyms.json        # Medical synonym ontology
│   ├── clinical_guidelines.json     # Clinical guidelines for safety checks
│   └── safety_rules.json            # Safety rules
├── agentos/                         # AgentOS framework (internal)
│   ├── agent/
│   │   └── agent.py                 # Base agent with reason-act loop
│   ├── memory/
│   ├── prompt/
│   ├── rag/
│   │   ├── store.py                 # ChromaDB vector store
│   │   ├── embedding.py             # BAAI/bge-base-zh embeddings
│   │   ├── bm25_retriever.py        # BM25 keywor