Incident learning faster

RCAgpt

An AI-assisted incident management tool that drafts RCA narratives from incident signals for faster review.

Outcome Roughly 5 hours to 10 minutes
Scale Incident signals, logs, notes, and post-mortem context

Problem

After firefighting an incident, engineers still had to reconstruct timelines, pull scattered signals together, and turn them into a consistent RCA draft.

Approach

Used LLMs and vector search to gather incident context, extract the likely timeline and contributing signals, then generate a draft for human review.

Contribution

Built and published the tool, shaped the review-first workflow, and focused the experience on reducing toil while keeping engineers accountable for the final RCA.

Measured outcome

Cut RCA preparation from roughly five hours of writing to about ten minutes of review and improved consistency across post-mortems.