PR reviews that slow your team down.

Every startup engineering team hits the same wall. Pull request reviews stack up. The reviewer with the most context becomes the bottleneck. Subtle issues — naming drift, missing null guards, half-applied patterns — slip through because the human reviewer is tired and the deadline isn't moving. The cost shows up later, as the bug report, the regression, the on-call page.

CodeSentinel sits between Git and Jira. When code lands on a watched branch, it picks up the diff, sends it through an LLM with a project-specific review prompt, and files structured Jira tickets for anything it flags. The human reviewer keeps the final say — they close what they disagree with. What CodeSentinel adds is consistency, speed, and the fact that it never goes on holiday.

• 01Push-triggered review pipeline — Git webhook → diff parser → LLM → Jira issue API.
• 02Project-specific prompt scaffolding so the rules match the codebase, not generic best-practice noise.
• 03Structured output schema — every finding maps to a Jira field. No free-form blobs in the ticket body.
• 04Idempotency on the commit SHA — re-running the same push doesn't duplicate tickets.
• 05Per-repo cost and rate-limit budget so it can be left running on a real engineering team.

Working today against a live product codebase — edugala.com — with a private repo and private Jira account. The point isn't the integration; it's the product thinking. What to flag, what to suppress, how the reviewer's workflow doesn't degrade. That's the signal.

• Node.js
• OpenAI API
• Git webhooks
• Jira REST API