SETUP_GUIDE.md

GAIA AI Agent - Hugging Face Space Setup

This directory contains an optimized GAIA AI agent designed for the Hugging Face Unit 4 final assignment.

🎯 Goal

Score 30+ on GAIA Level 1 questions to earn certification.

🚀 Quick Setup

1. Create a Hugging Face Space

1. Go to Hugging Face Spaces
2. Click "Create new Space"
3. Choose "Gradio" as the SDK
4. Upload all files from this hf_space directory

2. Set up API Keys

1. Get a free Groq API key from console.groq.com
2. (Optional) Get a Tavily API key from tavily.com
3. In your Space settings, add these as secrets:
   • GROQ_API_KEY: Your Groq API key
   • TAVILY_API_KEY: Your Tavily API key (optional)

3. Run the Evaluation

1. Open your Space
2. Login with your Hugging Face account
3. Click "Run Evaluation & Submit All Answers"
4. Wait for results (usually 2-5 minutes)

🧠 Agent Features

• Fast LLM: Uses Llama 3.1 70B via Groq for quick responses
• Web Search: Real-time information via Tavily API
• Math Tools: Built-in calculator for numerical problems
• Optimized: Streamlined for speed and accuracy
• Error Handling: Robust error management

📁 Files Overview

• app.py: Main Gradio application
• agent.py: Core GAIA agent implementation
• requirements.txt: Python dependencies
• system_prompt.txt: Agent instructions
• README.md: Space documentation
• .env.example: Environment variable template

🔧 Technical Details

The agent uses a multi-step approach:

1. Analysis: Determines if tools are needed
2. Tool Usage: Applies calculations or web search
3. Reasoning: Combines information for final answer
4. Formatting: Ensures proper "FINAL ANSWER:" format

🎯 Optimization for GAIA

• Focused on Level 1 questions (basic reasoning)
• Fast model selection (70B for capability, Groq for speed)
• Minimal tool overhead
• Direct answer extraction
• Error recovery mechanisms

📊 Expected Performance

Target: 30%+ accuracy on GAIA Level 1 questions

• Mathematical problems: High accuracy
• Web search questions: Good accuracy with Tavily
• Reasoning tasks: Moderate to high accuracy
• Overall: Should achieve certification threshold

🛠️ Customization

You can improve the agent by:

• Adjusting the system prompt
• Adding more specialized tools
• Fine-tuning the answer extraction
• Implementing caching mechanisms
• Adding more robust error handling

Good luck with your certification! 🎉