API Doctor

• 

  architecture diagram

Inspiration

APIs break in subtle ways: a tiny schema tweak can cascade into outages, client failures, and late‑night incidents. We wanted a system that catches risky version changes early, quantifies impact, and ships safe fixes before customers feel the pain.

What it does

API Doctor is an AI reliability agent that detects breaking API changes, estimates their blast radius, and proposes safe compatibility fixes before users are impacted. It combines schema diffs, runtime failures, and automated validation to generate a remediation plan and a PR with migration notes. The demo shows an end‑to‑end loop: failure detection → AI analysis → version‑impact preview → sandbox validation → fix PR.

Problem + impact

Breaking API changes silently cascade into production outages, failed client integrations, and expensive incident response. Teams often discover these regressions too late—after deployment and customer impact. API Doctor reduces downtime by catching risky schema changes early, quantifying impact, and shipping validated fixes with a migration plan.

How we built it

Key technical architecture + components Frontend (Next.js App Router): Dashboard and Inspector UI for failures, diffs, and AI fix plan.

Backend (Node.js + Express, TypeScript): Routes for failure capture, version impact analysis, and remediation workflows.

Buggy API repo: Intentionally broken endpoints + contract test to simulate schema regressions.

AI Analysis Engine (Gemini): Produces root cause, corrected payload, migration diff.

Validation Layer (Daytona sandbox): Replays schema tests or sample payloads to confirm fixes.

Challenges we ran into

Model availability and API versions for Gemini required model discovery and configuration. Daytona sandbox storage limits forced cleanup before rerunning fixes. Keeping the UI realtime while switching from mock to live signals without breaking the flow.

Accomplishments that we're proud of

Full end‑to‑end loop: error → analysis → fix → test → PR → review. Automated PR generation from a Daytona sandbox. Live config/status panel and Inspector data tied to actual backend state.

What we learned

Strong error context (Sentry + schema) makes AI remediation far more accurate. Daytona enables safe, reproducible validation in seconds. Integrations matter as much as the AI—closing the loop is the real win.

What's next for API Doctor

Real schema diff parsing from GitHub PRs. Automated compat‑layer generation and staged rollouts. Multi‑agent remediation for parallel fixes across services. Deeper CI/CD integration for automatic gating and rollback.

Built With

• coderabbit
• daytona
• sentry