Autonomous

**Autonomous AI Software Engineer — Human-Guided AI Code Repair** Autonomous AI Software Engineer is a local AI-powered development assistant that helps developers identify Python bugs, generate focused code patches, review proposed changes, and validate fixes through automated testing. The project addresses three common challenges in software development: understanding faulty code, verifying AI-generated modifications, and safely testing code changes. **How it works:** 1. Developers upload a Python file or select a bundled sample. 2. FastAPI performs static analysis using Python's Abstract Syntax Tree (AST) without executing uploaded code. 3. Meta Llama 3.2 3B, running locally through Ollama, analyzes the code and proposes a focused patch. 4. A second Llama inference call reviews the proposed modification. 5. The application generates a unified diff, allowing developers to inspect and explicitly approve the exact change. 6. SHA-256 verification binds approval to the reviewed source and patch. 7. Approved patches can be downloaded. Reviewed bundled samples can also be validated through Pytest inside restricted Docker containers. **Technology stack:** React, Vite, Tailwind CSS, Python, FastAPI, Meta Llama 3.2 3B, Ollama, Docker, and Pytest. **Verified results:** The prototype passed 20 backend tests. In the live parity demonstration, the original code failed five tests, while the approved patch passed all five. **Safety:** Uploaded code is never executed. Container testing is restricted to predefined, reviewed samples. The system requires human approval before downloading patches. This project demonstrates how locally hosted Meta AI models can support a transparent, human-controlled software repair workflow. The current MVP focuses on individual Python files rather than unrestricted autonomous repository modification. **GitHub:** https://github.com/Edward2106/Autonomous-AI-Software-Engineer