1 of 11

Self-Healing Bug Agent

An autonomous AI engineer that fixes GitHub issues before opening a Pull Request.

ASMAE SERJI GHIYAS MEHMOOD XIANGYI JING ARAFAAT HUSAIN

2 of 11

Software bugs cost developers time

The Bottleneck

Developers spend hours debugging.

AI can generate code...

...but cannot prove that the fix actually works.

The Flaw

Current AI coding assistants are open-loop. They suggest fixes. They don't verify them.

Maybe works

3 of 11

Closed-Loop Self-Healing

The AI never decides it succeeded.

Only deterministic verification allows a Pull Request.

4 of 11

System Architecture

BACKEND

Python

FastAPI

AI

GPT-5.6

Codex

FRONTEND

React

Tailwind CSS

VERIFICATION

Git

Pytest

5 of 11

Modules

Backend & Orchestrator

Coordinates the entire autonomous workflow.

AI Repair Agent

d

Analyzes bugs and creates intelligent patches.

Sandbox & Verification

Proves the patch works in an isolated environment.

Interactive Dashboard

Visualizes everything happening in real-time.

6 of 11

Safe Verification Before Every Pull Request

No patch can reach GitHub without passing verification.

Isolated execution

Retry mechanism (5 max)

Regression testing

Structured reports

Adapter with orchestrator

7 of 11

Live Workflow States

01

Preparing Workspace

02

Reproducing Issue

03

Diagnosing Root Cause

04

Writing Patch

05

Writing Test

06

Executing Verification

07

Ready for PR

08

PR Created

8 of 11

Demo Scenario

9 of 11

Why our approach is different

Traditional AI

Self-Healing Bug Agent

Suggests code

Executes code

No validation

Sandbox verification

No regression tests

Generates regression tests

Manual review

Verified PR

Open loop

Closed loop

10 of 11

Production Roadmap

Building trustworthy autonomous software engineers.

Docker isolation

GitHub App

Live streaming dashboard

PostgreSQL persistence

Multi-language support

Kubernetes execution

Parallel verification

11 of 11

Tank You