Skip to content

pegasi-ai/feather

Repository files navigation

🪽 Pegasi Feather

License Python 3.7+ Code style: black

✨ AI Testing Framework

Feather is a lightweight framework for statistical testing and validation of LLM outputs and behaviors. With Feather, you can implement comprehensive test suites, automated evaluations, and behavioral checks to ensure your AI applications perform reliably and align with specified requirements.

🔍 Core Features

  • 📊 Statistical Testing: Comprehensive testing suite for model behavior validation
  • ✍️ Evaluations: Quantitative and qualitative metrics to measure model performance
  • 🛡️ Validations: Simple safety checks and output validation

⚡ Get Started

  1. Grab your API key here: app.pegasi.ai
  2. Quickstart Evals notebook: Open In Colab
  3. DeepSeek-R1 on FinQA notebook: Open In Colab

💼 Roadmap

  • Establish AI validators
  • Setup out-of-the-box Judges
  • Add distribution-based testing
  • Expand statistical validation tools
  • Improve test results visualization
  • Enable custom test case creation
  • Add community-driven test suites