🎯 Grasper - AI-Powered Data Analysis Agent

Grasper is a production-ready AI data analysis system built with PydanticAI, Google Gemini 2.0 Flash, and functional programming principles. It provides a complete workflow from questions to executable code with memory management, error recovery, and real-time monitoring.

🌟 Key Features

🤖 Complete AI Workflow

• Questions → Tasks: Breaks down complex questions into executable tasks
• Code Generation: Uses Google Gemini 2.0 Flash via PydanticAI to generate Python code
• Execution: Safely executes code using subprocess with timeout and error handling
• Error Recovery: Automatically detects and fixes code errors in a loop until successful
• Memory Management: Maintains workflow state and results throughout the process

🛠 Technical Excellence

• Real-time Monitoring: Logfire integration for observability and debugging
• Concurrent Processing: Parallel task execution with throttling and rate limiting
• Type Safety: Full type hints with Pydantic models
• Error Resilience: Comprehensive error handling and recovery mechanisms

🚀 Quick Start

1. Setup Environment

# Install dependencies with uv
uv sync

# Create environment file from template
cp .env.example .env

2. Configure API Keys

Edit .env file:

# Required: Google API Key for Gemini
GOOGLE_API_KEY=your-google-api-key-here

# Optional: Logfire for monitoring
LOGFIRE_TOKEN=your-logfire-token

# Optional: Environment settings
ENVIRONMENT=development
DEBUG=true

Get Google API Key:

1. Go to Google AI Studio
2. Create a new API key
3. Add it to your .env file

3. Start the Server

# Development server with hot reload
uv run uvicorn api.complete_api:app --host 0.0.0.0 --port 8000 --reload

📋 Usage Examples

1. Process Data Analyst Input File

Add your data analyst input to data_analyst_input.txt, then:

curl "https://app.example.com/api/" -F "data_analyst_input.txt=@data_analyst_input.txt" -F "image.png=@image.png" -F "data.csv=@data.csv" 

🏗 Architecture

Data Analyst Input → Task Breakdown → Code Generation → Execution → Error Recovery → Results

Core Components:

• agents/: AI agents using PydanticAI with Google Gemini
• core/: Workflow engine, functional utilities, type definitions
• api/: FastAPI endpoints with real-time monitoring
• monitoring/: Logfire integration and observability

📊 Monitoring & Observability

• Logfire Integration: Real-time logs, traces, and metrics
• Performance Metrics: Request rates, response times, error rates
• AI Analytics: Token usage, model performance, success rates

Access metrics at: http://localhost:8000/metrics

99% code written by llm

---

Built with ❤️ using PydanticAI