Resurrecting Li Bai with Cypher Queries and LLM APIs

What it does

meet-libai wraps a Neo4j knowledge graph of the Tang poet Li Bai inside a Gradio chat interface, using LangChain-style RAG to answer structured questions about his life and works via Cypher-backed retrieval. It also pipes requests to third-party Chinese LLMs for general chat, text-to-image, and text-to-speech generation, plus a poetry mini-game.

The interesting bit

The project treats a specific historical figure as a first-class domain for retrieval-augmented generation, binding classical scholarship to modern APIs rather than building yet another generic chatbot. The result is less an AI poet and more a digital docent that happens to run on PyTorch and Cypher.

Key highlights

• Neo4j graph database with explicit Cypher schemas for Li Bai’s relationships, works, and historical context.
• RAG pipeline that distinguishes relationship queries, attribute lookups, and open-ended LLM chat.
• Multimodal output via edge-tts and ZhipuAI for voice narration and image generation.
• Interactive poetry game (飞花令) included as a cultural engagement demo.
• Configurable LLM backends including ZhipuAI GLM-4, Moonshot Kimi, and local Ollama instances.

Caveats

• No complete dataset is provided; the repo contains only sample Cypher nodes due to copyright restrictions, so you must supply your own curated Li Bai data.
• Requires external API keys (ZhipuAI is mandatory for image generation), so it does not run fully offline out of the box.

Verdict

A useful reference if you are prototyping Chinese-language cultural RAG bots or experimenting with Neo4j-to-LLM bridges. If you need a turnkey digital humanities archive with full open data, this is just a scaffold.