feat: add ShenNongMi — TCM Knowledge Graph RAG with LangGraph + Neo4j

[Happy-Chen-CH]

Description

Add ShenNongMi (神农觅 — "The Divine Farmer Seeks"), a self-contained knowledge-graph RAG tutorial for Traditional Chinese Medicine Q&A, built per the feedback on #1066.

Architecture

User query
  → Intent classification (RoBERTa+LoRA, local model)
  → Entity extraction (LLM)
  → FAISS entity matching (BGE-Large-Zh-v1.5)
  → Cypher generation with self-correction loop (max 3 retries)
  → Neo4j execution
  → Natural-language answer (LLM)

Changes vs. original submission (#1066)

Feedback	Action
Full multi-service product	✅ Slimmed to the LangGraph + Neo4j Q&A core only (41 files, ~432 KB)
ToS-violating auto-publishing	✅ Removed Playwright/Xiaohongshu pipeline entirely
13.6 MB screenshots	✅ Replaced with 2 placeholder PNGs (140 bytes total)
Invalid model default	✅ deepseek-chat (valid DeepSeek model)
pip install broken	✅ Flat, semantic directory names; requirements.txt installs cleanly
MIT inside Apache-2.0 repo	✅ Apache-2.0 licensed

What's included

• 8-node LangGraph workflow: intent → extract → FAISS match → Cypher gen → validate → execute → answer (+ non-TCM fallback)
• Neo4j knowledge graph with 6 entity types and 6 relationship types (schema in kg_setup/tcm_metadata.json)
• Streamlit frontend (Q&A + graph explorer) with FastAPI SSE streaming backend
• Integration tests with mocked LLM/Neo4j/FAISS (run: python langgraph_workflow/langgraph_more_nodes.py)
• Knowledge graph setup scripts (Neo4j import + FAISS index builder)

Run it

cd rag_tutorials/shennongmi_tcm_knowledge_graph
pip install -r requirements.txt
cp .env.example .env  # add your DeepSeek API key + Neo4j credentials
python fastapi_app/main.py            # terminal 1
streamlit run streamlit_app/app.py    # terminal 2

Related

• Resolves [New App] ShenNongMi — TCM Knowledge Graph RAG with LangGraph + Neo4j + DeepSeek #1065
• Follow-up to feat: add ShenNongMi — TCM Knowledge Graph RAG with LangGraph + Neo4j + DeepSeek #1066 (closed)