# Rag Neural

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Harishwaran K](https://cerebralvalley.ai/u/Harish_22)
- **GitHub:** https://github.com/HARIScyber/PatchPilot-.git
- **Demo video:** https://youtu.be/S-o5Sv6azW8?si=BPJwwUjh1utM8AC8
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1158

# Project Name: PatchPilot

## Project Description

PatchPilot is a self-verifying AI software engineering agent designed to automate the software debugging and repair lifecycle.

Traditional AI coding assistants primarily generate code suggestions, leaving developers responsible for identifying root causes, creating regression tests, executing them, and determining whether a fix actually works.

PatchPilot takes a different approach. It analyzes repository structures, performs AST-based static inspection, identifies potential defects, develops repair plans, generates minimal code patches, and creates regression tests designed to reproduce the reported bug.

Its verification architecture evaluates patches against defined tests and uses execution evidence to distinguish verified fixes from unverified suggestions. When test failures occur, the system is designed to support bounded repair iterations using actual failure feedback.

Key capabilities include:
- Repository ingestion and source-code exploration.
- AST-based static defect inspection.
- AI-powered root-cause analysis and repair planning.
- Minimal patch and diff generation.
- Regression test generation.
- Verification evidence and test reporting.
- Human approval before applying patches.
- Provider abstraction for integrating different language models.

The central innovation is moving beyond code generation toward evidence-driven autonomous software repair.

PatchPilot aims to reduce debugging effort, improve software reliability, and make AI-generated code changes more accountable and testable.

Current public deployment includes an explicitly labelled demo mode. Live model inference and secure code execution remain separate production-readiness requirements.

## More from Meta Global AI Developer Hackathon

- [BidLab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1155)
- [Amr Hassan](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1156)
- [Jepa as a Tool](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1157)
- [CukupAI](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1159)
- [SIERA Labs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1160)
- [Atlas-by-rookie](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1161)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1158. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt