Summary This project, "Kemo" (可萌), tackles the complexity of the Pokémon universe's vast knowledge base by creating a specialized, multimodal AI assistant. It targets Pokémon fans and developers seeking a transferable template for domain-specific AI solutions. The system constructs a knowledge graph from sources like Wikipedia and Baidu Tieba, integrating it with LLMs via LangGraph and GraphRAG. This enables precise natural language Q&A, visual graph exploration, and unique features like real-world map integration for Pokémon locations, offering an interactive and informative experience.

How It Works The system builds a detailed Pokémon knowledge graph using data from Wikipedia and Baidu Tieba, enhanced by automated NER annotation (roberta + TF-IDF + rule matching). It employs LangGraph to orchestrate complex reasoning, combining GraphRAG with web search and a knowledge base. A fine-tuned, domain-specific LLM ("Kemo") powers the conversational interface, allowing accurate answers on intricate lore like evolution paths and type matchups, alongside visual exploration and location-based services.

Quick Start & Requirements

• Installation: Clone the repository, configure environment variables (src/.env, settings.py), install Python dependencies (pip install -r requirements.txt), and run Docker Compose for core services (Neo4j, Milvus, Whisper, MySQL). Data import scripts (import_graph.py, import_pokemon_map.py) are required. Backend (FastAPI), MCP server, and frontend (Vue.js) services must then be started.
• Prerequisites: Docker/Docker Compose, Node.js ≥ 18, Python ≥ 3.11.
• Access: http://localhost:3100/.

Highlighted Details

• Fine-tuned domain-specific LLM ("Kemo").
• Knowledge graph construction from Wikipedia/Baidu Tieba.
• LangGraph integration for GraphRAG and agent orchestration.
• Real-world map integration for Pokémon locations.
• Multi-modal search support (KG, web, KB, MCP, voice).

Maintenance & Community The provided README does not detail maintainers, community channels, or a roadmap.

Licensing & Compatibility Released under the MIT License, allowing free commercial and personal use, provided original author and source information is retained for derivative works.

Limitations & Caveats No explicit limitations are listed. The setup involves multiple services and data import steps, indicating a potentially complex deployment. The project is positioned as a template, requiring further domain-specific adaptation.