AI Reliability Copilot

What it is

An AI incident-response copilot for reliability and SRE workflows. Paste raw incident context from logs, metrics, on-call notes, or alert JSON, and the app returns a structured nine-section response covering summary, severity, root-cause hypotheses, investigation checklist, mitigation, postmortem, and follow-ups.

Key Features

• Structured incident analysis — converts noisy alert context into a consistent reliability report.
• Alert JSON parsing — supports operational inputs from tools like Datadog, PagerDuty, and Sentry.
• Scenario library — includes sample incidents for DB pool exhaustion, OOM crashloops, and checkout error spikes.
• Bilingual output — supports English and Chinese triage workflows.
• Eval surface — exposes prompt versions and reliability scenarios for comparing output quality.
• Production demo controls — rate-limited demo flow designed for public deployment.

Technical Focus

The product is shaped around high-pressure incident work: keep the input path fast, preserve raw context, and force the model into an operationally useful structure instead of a generic explanation. The output format makes it easier for an on-call engineer to separate known facts, likely hypotheses, immediate mitigations, and post-incident follow-ups.

What I Learned

• Incident copilots need structure more than verbosity; responders need the next useful action.
• Prompt evals are easier to reason about when scenarios mirror real alert streams.
• Bilingual reliability tools need domain-specific phrasing, not just translation.

Tech Stack

Next.js, TypeScript, Vercel, AI SDK, structured prompting, prompt evals, incident-response workflows