Revund.dev
AI code review that reads PRs like a senior engineer.
3 followers
AI code review that reads PRs like a senior engineer.
3 followers
AI code review that reads pull requests like a senior engineer — structured findings with severity, location, and the why behind every flag. Confidence-gated, deterministically fingerprinted, audit-logged. TypeScript today, polyglot next.
Hey Product Hunt 👋
Most AI code reviewers feel like black boxes — you get a flag, no rationale, and no way to make it stop coming back next PR. We built Revund to fix that.
Three layers under the hood:
🔹 Signal Quality — six inputs (diff, files, resolved symbols, tsc diagnostics, PR metadata, repo config) normalize into one deterministic ContextBundle. Same PR in, same bundle out.
🔹 Review Engine — four specialist passes run in parallel: security, performance, architecture, style. Each has its own prompt, its own schema, emits its own findings.
🔹 Trust Mechanics — every finding passes three gates: confidence ≥ 0.6, a deterministic fingerprint, and the team's suppressions dict. Suppress it once, it stays suppressed across every future review.
Every finding ships with severity, location, body, and a required "why" — the rationale we make the model commit to before surfacing anything. No why = no finding.
TypeScript today, polyglot next. CLI first, GitHub/GitLab PR bot , self-hosted runner for enterprise after.
Would love your feedback — especially: what would make you actually trust an AI reviewer on your repo?
— Wes Guirra