CodeGuard AI

CodeGuard AI is an AI-assisted code health and verified debugging platform designed for software engineering teams maintaining existing codebases. Modern development teams spend significant time deciding where to investigate, understanding the root cause of bugs, validating AI-generated fixes, and rediscovering solutions that were already found in the past. CodeGuard AI connects these activities into one workflow. First, CodeGuard scans a repository and produces a transparent heuristic risk estimate for each source file using measurable signals such as code complexity, Git churn, and test-coverage gaps. The dashboard separates predicted risk from verified incident evidence so teams can understand why a file is considered risky without treating the score as a prediction or guarantee of failure. When a risky area contains a reproducible bug, CodeGuard provides an investigation workflow that identifies the suspected problem, root cause, supporting code evidence, and a candidate fix. The important difference is that the proposed fix is not automatically trusted. CodeGuard independently verifies the fix using an explicit red-to-green workflow: the regression test must fail against the broken implementation, pass against the proposed fix, and the existing regression suite must also pass. Only after these conditions succeed can the incident be saved as verified knowledge. Verified incidents are stored in persistent Verified Memory with the problem, root cause, fix, regression test, repository context, and verification evidence. When a similar issue appears later, CodeGuard can surface previous verified incidents as debugging context and a fix candidate. Previous knowledge is never treated as automatic truth; the current code must be independently re-checked and verified again. The system also identifies “Cold Zones”: areas with elevated estimated risk but no verified incident evidence and no completed investigation. This helps teams distinguish “no known verified incident” from “low risk.” The target users are software developers, engineering teams, technical leads, and operations teams responsible for maintaining existing applications. The workflow is designed to reduce debugging effort, repeated investigation, and unverified AI-generated changes while building organizational knowledge over time. The project is built as a practical local prototype using Python/FastAPI for the backend, React/Vite for the dashboard, SQLite for verified memory, Git history and static analysis for risk signals, and automated pytest-based verification. The result is a continuous loop: find risk → investigate → verify → remember → safely reuse → update risk evidence.