Ship the AI Code Reviewer

the end-to-end review pipeline with a regression check

Part of: AI Code Reviewer

You have every piece. A schema and validator, a language detector, a static pre-pass, numbered prompts, a merge step, security rules with a gate, and an evaluation. Shipping means wiring them into one path from a paste to a report, and adding the check that keeps the reviewer from getting quietly worse. Finish this lesson and the build lands in your Portfolio . The pipeline. Order matters. Detection picks the rule table. The pre-pass runs before the model, so a failed or slow model call still leaves the user with real findings. The model's reply goes through the validator before it touches anything. Merge runs last, over both sources. A small endpoint. The front end is a textarea, a Review button, and a list that renders each finding with a severity color and a link to its line. Cap the input size on the server so a huge paste can't run up a bill. Keep it from getting worse. The evaluation from lesson 7 becomes a regression check . Store the recall from the last version you were happy with as a baseline. Every time you change the prompt or the rules, run the labeled set again. If recall drops below the baseline, the check fails and the change doesn't ship. Run it in CI next to your

Challenge: Review and Guard