RuleMindLLM

RuleMindLLM is a local language model built using Ollama and ChromaDB that answers rule-based questions about popular games like Uno, Monopoly, and Poker.

It uses Retrieval-Augmented Generation (RAG) to provide accurate, grounded responses by leveraging vector similarity search on embedded chunks of official rulebooks.

---

🧠 Tech Stack

🔹 Ollama (using the Gemma model)

• Runs the local LLM efficiently on your machine
• Handles natural language understanding and response generation
• Accepts external context passed from ChromaDB to ground responses in documents

🔹 ChromaDB

• Converts game rule PDFs into vector embeddings
• Stores embeddings for fast similarity search
• Retrieves the most relevant rule chunks based on user queries
• Supports RAG: provides context to the LLM at inference time

---

📚 Example Use Case

Ask:
"Can you stack a Draw Two in Uno?"

→ RuleMindLLM will return an answer directly from the Uno rulebook PDF, explaining whether stacking is officially allowed.

---

🚀 How to Run (WIP)

1. Install Ollama and run ollama run gemma
2. Clone this repo
3. Load game rule PDFs → embed → store in ChromaDB
4. Ask questions via CLI or UI (under development)

---

💡 Motivation

LLMs are great at language, but they often hallucinate. RuleMindLLM anchors LLMs in real documents, making them practical assistants for rule-based domains like games.