Local AI Code Review Lab

A Spring Boot portfolio project for testing local AI code reviews against fake pull requests with known bugs.

3 fake PRs with known bugs

3/3 expected bugs caught

0 private repos or real tickets

Why This Project Exists

A local model can review code, but a good AI engineer should test the result. This project creates fake PRs with known bugs, asks a local model to review them, and scores whether the review found the real issue.

Review Cases

Case	Risk	Expected finding
`PR-001`	authorization bypass	Ticket close endpoint skips owner check.
`PR-002`	validation	Discount percent accepts negative or above-100 values.
`PR-003`	secret exposure	Payment client logs an Authorization bearer token.

First Local Run

Case	Score	Grade
`PR-001`	1.0	pass
`PR-002`	1.0	pass
`PR-003`	1.0	pass

Honest limitation: the local 27B model is useful, but it is not instant. Future runs should compare speed, quality, and output cleanliness across models.

Local Workflow

./mvnw spring-boot:run
scripts/run_review.sh PR-001
scripts/score_review.sh PR-001 results/PR-001-review.md

Portfolio Story

Built a Spring Boot evaluation app for local AI code review using fake PRs with known security and validation bugs, REST scoring endpoints, a Thymeleaf dashboard, and a Hugging Face-ready static report.