Technical Overview of the Collective Governance GPT Bot

[nutkung1]

Hi, Optimism community,

I'm Nut from @catalyzt-tech, and I’ve been working on the technical solution for the Collective Governance GC Bot. Here's a breakdown of the current model's architecture and its key technical features:

Key Features:

• Comprehensive data coverage: Pulls information from multiple data sources, including:
  • Governance documents
  • Forum posts and discussions
  • The Charters, Law of Chains, and other governance-relevant documents
  • GitHub repositories and technical discussions
• Updatable Knowledge Base: Automatically updated with new governance information to ensure the chatbot is always up to date.
• Graph-based knowledge representation: Enhances the chatbot’s ability to understand and respond to complex queries by mapping relationships between governance topics.
• Source validation and referencing: Provides references and source links for every answer to ensure users can verify the provided information. (Available on the web)
• Embeddable widget: The bot can be easily integrated into different platforms, alongside a standalone website for wider accessibility.

Technical Details:

• Model: Meta llama 3.1 70b
• Multi-agent system: Can answer questions related to Optimism governance, leveraging both dense (vector-based) and sparse (keyword-based) search. Also, multi-agent architecture can check for model response hallucination and provide a clearer and more concise response to users with low latency.
• Efficient hybrid search architecture: Combines Cohere embeddings for dense search and bm25 for sparse embeddings to ensure accurate retrieval of governance documents, forum posts, GitHub discussions, and other sources.
• Optimized performance: Designed to handle large document collections while maintaining fast response times using state-of-the-art retrieval techniques.
• Multi-threading: Designed to handle large volumes of requests from users without bothering the performance and quality of response.

Our product will be available on Websites, Discord, and Telegram

Looking for Feedback: I would love to hear your thoughts on:

• Additional data sources we should consider
• Ideas to improve accuracy and interaction
• Suggestions for keeping the knowledge base updated efficiently

We looking forward to collaborating and making governance more accessible for everyone!

Best regards,
@nutkung1 (@catalyzt-tech)
@Gunn-Wangwichit (@catalyzt-tech)