Skip to content

Developer tooling

AI PR Reviewer

Automated code review for pull requests on GitHub and Azure DevOps, with inline comments and summaries from OpenAI, Claude or Gemini.

Discuss a similar project
My role
Design and engineering
Stack
TypeScriptNode.jsExpressDockerVitest

The problem

Review queues slow teams down, and large diffs are hard for LLMs to handle reliably. The tool has to be configurable, quiet and safe to run in CI.

What I built

  • Runs three ways: as a GitHub Action, a GitHub App, or a webhook server for GitHub and Azure DevOps.
  • A provider abstraction so the same review pipeline works with OpenAI, Anthropic Claude or Google Gemini.
  • Smart filtering of lock files, generated code and binaries, chunk splitting for large diffs, and severity levels (critical, warning, suggestion) with a configurable threshold.
  • Production hygiene: a token-bucket rate limiter, exponential backoff retries, per-repo YAML rules and custom instructions.

How it fits together

  1. PR event
  2. Diff filter + chunking
  3. LLM provider adapter
  4. Severity filter
  5. Inline comments + summary

Highlights

  • Three deployment modes, two platforms, three LLM vendors
  • Rate limiting and retries built in
  • Configurable per repository