Description:
The Autonomous Project Reviewer Agent is a compact, explainable agentic system designed to automatically review GitHub repositories and generate human-readable summaries, scores, and actionable recommendations. Unlike typical AI code reviewers that rely heavily on LLMs, this system prioritizes deterministic static analysis for correctness and uses LLMs only for high-level synthesis.
What it does:
- Safely clones a GitHub repository and enumerates files
- Reads only whitelisted files with strict size and extension checks
- Runs deterministic checks for:
- Missing docstrings
- TODO/FIXME markers
- Oversized or poorly structured files
- Scores the project across:
- Code Quality
- Documentation
- Structure & Tests
- Uses an LLM (Groq) to:
- Summarize the project in plain English
- Generate clear, actionable improvement suggestions
- Outputs results via:
- CLI
- Streamlit UI
- Exportable Markdown / PDF reports
Why this project matters:
Most AI reviewers hallucinate or over-generalize. This project demonstrates a production-style agentic approach where:
- Core evaluation is deterministic and explainable
- LLMs are used only where they add value (synthesis, communication)
- Outputs are reliable, auditable, and resume-ready
Design Philosophy:
- Deterministic logic > probabilistic guesses
- Explainability over black-box outputs
- Agentic systems that are:
- Reliable
- Practical
- Reviewable
- Real-world usable
Tech Stack:
- Language: Python
- UI: Streamlit
- LLM: Groq API
- Tooling: GitPython, python-dotenv
Project Status:
- Core system complete and functional
- CLI + UI supported
- Roadmap includes GitHub Actions and multi-repo benchmarking