---
title: GAIA Benchmark Agent
emoji: 🧠
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 5.25.2
app_file: app.py
pinned: false
hf_oauth: true
hf_oauth_expiration_minutes: 480
---

# GAIA Benchmark Agent

This Hugging Face Space hosts a GAIA (General AI Assistant) benchmark agent designed to solve certification challenges across various domains of AI and machine learning.

## Features

- Processes questions from the GAIA benchmark
- Uses LangChain and OpenAI's language models
- Analyzes questions and identifies their types
- Retrieves relevant context when needed
- Generates accurate, well-reasoned answers
- Integrates with external information sources:
  - SerpAPI for real-time web search capabilities
  - YouTube for video content search and transcript analysis
  - Tavily for AI-optimized search results
  - Audio processing for speech-to-text conversion and analysis

## Usage

1. Log in to your Hugging Face account using the button
2. Click 'Run Evaluation & Submit All Answers' to:
   - Fetch questions from the GAIA benchmark
   - Run the agent on all questions
   - Submit answers and see your score

## Implementation Details

The agent uses a modular architecture with specialized handlers for different question types:
- Factual knowledge questions
- Technical implementation questions
- Mathematical questions
- Context-based analysis questions
- Ethical/societal impact questions
- Media content questions (videos, podcasts, audio recordings)
- Current events questions
- Categorization questions with enhanced botanical classification

### Botanical Classification

The agent has been enhanced with comprehensive botanical classification capabilities, allowing it to:
- Accurately distinguish between botanical fruits and vegetables
- Provide detailed explanations of botanical classifications
- Correctly identify commonly misclassified items (tomatoes, bell peppers, cucumbers, etc.)
- Explain the difference between botanical and culinary classifications

### External Information Sources

The agent can access external information to provide more accurate and up-to-date answers:

- **SerpAPI Integration**: Enables real-time web search capabilities for current events and factual information
- **YouTube Integration**:
  - Search for relevant videos on specific topics
  - Extract and analyze video transcripts for information
- **Tavily Search**: AI-optimized search engine that provides relevant results for complex queries

### Audio Processing Capabilities

The agent has been enhanced with audio processing capabilities, allowing it to:
- Transcribe audio files using OpenAI's Whisper API with Google Speech Recognition fallback
- Extract ingredients from recipe audio recordings
- Process and analyze spoken content from various audio formats
- Format responses according to user requests for audio content

### API Keys Configuration

To use the external information sources, you need to set the following API keys in your environment:
- `SERPAPI_API_KEY`: For web search capabilities
- `YOUTUBE_API_KEY`: For YouTube video search and transcript analysis
- `TAVILY_API_KEY`: For AI-optimized search results
- `WHISPER_API_KEY`: For audio transcription (defaults to OPENAI_API_KEY if not set)

## Repository

The code for this agent is available at: https://huggingface.co/derkaal/GAIA-agent

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference