~94s
Full pipeline latency
PR Pilot autonomously reviews pull requests with a five-agent pipeline — Reviewer reviews, Fixer patches, Tester validates, Verifier gates, Escalator decides. A full review lands in ~94 seconds for about four cents.
A FastAPI webhook triggers AgentChain orchestration over the Gemini API; results post back to the PR. The Fixer sees what the Reviewer found, the Verifier checks the Fixer's patches, and the Escalator makes the final call.
Zero-downtime, auto-scaling, request concurrency — one gcloud run deploy. Scales linearly: 100 PRs for under $4.
Every agent's reasoning, every generated fix, every decision logged as structured JSON to data/reviews/.
Gemini API, Google Cloud Run, and Stripe — competition rules met, with documentation to match.
PR Pilot operates at three levels, each escalating only when confidence demands it — the safety gate that separates autonomous review from automated approval.
Low-confidence findings posted as review comments with fixes attached. Nothing is blocked — the author sees what to change and why.
Medium-confidence issues that should gate the merge. Fixes provided inline; the PR stays blocked until the author responds.
High-confidence, fully-clean PRs are approved autonomously. The escalator only reaches this level when every gate passes.