MIRA v0 Prototype 🧬

AI-Powered Knowledge Assistant for DeSci & Molecule

MIRA (Molecule Insights Research Assistant) is an experimental prototype that demonstrates how modern AI can intelligently provide accurate information about decentralized science (DeSci) and Molecule's ecosystem.

🎯 What is MIRA?

MIRA is a smart chatbot that can answer questions about anything related to Molecule and the DeSci ecosystem. This includes, but is not limited to:

• DeSci (Decentralized Science) - The movement to make scientific research more open and accessible
• Molecule Platform - All features, products, and services offered by Molecule
• IP-NFTs - Tokenizing intellectual property and research assets
• Research Funding - Grants, investments, and new funding mechanisms
• DAOs & Governance - BioDAOs, VitaDAO, and decentralized organizations
• DeSci Tools - Various tools and protocols in the ecosystem
• Community & Events - DeSci conferences, initiatives, and collaborations
• And much more! - Any topic connected to Molecule's mission and the broader DeSci movement

Key Features

• 🤖 Smart Decision Making: MIRA decides on its own whether it has enough information to answer your question
• 📚 Local Knowledge Base: Contains curated information about Molecule and DeSci
• 🌐 Web Search Fallback: Automatically searches the web when local knowledge isn't enough
• 👍👎 Feedback System: Help improve MIRA by rating responses
• 🔍 Source Transparency: Always shows where information comes from

🏗️ How It Works

MIRA uses a three-step intelligent process called "Agentic RAG" (Retrieval Augmented Generation):

1. Search Local Knowledge

When you ask a question, MIRA first searches its local knowledge base for relevant information using advanced semantic search that understands meaning, not just keywords.

2. Evaluate & Decide

MIRA then evaluates the information it found:

• Sufficient Context? → Use local knowledge to answer
• Insufficient but Relevant? → Search the web for current information
• Off-topic? → Politely decline with an explanation

3. Generate Response

Based on its decision, MIRA either:

• Provides an answer from trusted local sources
• Searches the web for up-to-date information (focusing on trusted domains)
• Explains that the question is outside its expertise

🛠️ Technology Stack

MIRA combines several technologies:

Chainlit (User Interface)

• Responsive chat interface
• Real-time streaming responses
• Interactive feedback buttons

LanceDB (Knowledge Storage)

• Vector Search: Finds information by meaning and context
• Hybrid Search: Combines semantic understanding with keyword matching
• Fast & Efficient: Delivers relevant results quickly

OpenAI GPT-4o (AI Brain)

• Understands natural language questions
• Evaluates context quality
• Generates human-like responses
• Web search capabilities for current information

Langfuse (Monitoring & Analytics)

• Tracks system performance
• Collects user feedback
• Helps improve responses over time
• Ensures quality and reliability

📊 The Workflow

Here's how MIRA processes your questions:

┌─────────────────────┐
│   User Question     │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│  Search Local       │
│  Knowledge Base     │
│  (LanceDB)          │
└──────────┬──────────┘
           │
           ▼
┌─────────────────────┐
│  Evaluate Context   │
│  Is it sufficient?  │
└──────────┬──────────┘
           │
    ┌──────┴──────┬─────────────┐
    │             │             │
    ▼             ▼             ▼
┌─────────┐ ┌─────────┐ ┌─────────┐
│   Yes   │ │   No,   │ │   No &  │
│         │ │   but   │ │  Off-   │
│  Local  │ │Relevant │ │  topic  │
│ Answer  │ │         │ │         │
└─────────┘ │   Web   │ └─────────┘
            │ Search  │
            │ (GPT-4o)│
            └─────────┘
                │
    ┌───────────┴─────────────┐
    │                         │
    ▼                         ▼
┌─────────────────────┐ ┌─────────────────────┐
│  Generate Answer    │ │ "Sorry, I can't     │
│  with Sources       │ │  help with that"    │
└──────────┬──────────┘ └──────────┬──────────┘
           │                       │
           └───────────┬───────────┘
                       │
                       ▼
            ┌─────────────────────┐
            │   Display Answer    │
            │   👍    👎          │
            └─────────────────────┘
                       │
                       ▼
            ┌─────────────────────┐
            │ Langfuse Monitoring │
            │ (Track & Improve)   │
            └─────────────────────┘

The Decision Process Explained:

1. You ask a question → MIRA receives your query
2. Searches local knowledge → Uses LanceDB's semantic search to find relevant information
3. Makes a smart decision:
   • ✅ Sufficient info? → Answers from local knowledge
   • 🌐 Not enough but relevant? → Searches the web for current info
   • ❌ Off-topic? → Politely declines
4. Generates response → Creates a helpful answer with sources
5. You rate the answer → Helps MIRA improve over time

🚀 Getting Started

Prerequisites

• Python 3.8 or higher
• Virtual environment (recommended)

Installation

1. Clone the repository

   git clone [repository-url]
   cd molecule-chainlit

2. Create and activate virtual environment

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

4. Set up environment variables Create a .env file with:

   OPENAI_API_KEY=your_openai_api_key
   LANGFUSE_PUBLIC_KEY=your_langfuse_public_key
   LANGFUSE_SECRET_KEY=your_langfuse_secret_key
   

5. Run the application

   chainlit run app.py

6. Open your browser Navigate to http://localhost:8000

⚠️ Important Notes

This is a Prototype

• MIRA v0 is an experimental prototype, not a production-ready application
• Used for learning about robust AI systems and gathering user feedback
• Responses should be verified for critical decisions

Data Sources

• Local knowledge base contains curated information about Molecule and DeSci
• Web search results are filtered to prioritize trusted domains like:
  • molecule.to
  • molecule.xyz
  • bio.xyz
  • vitadao.com
  • ...