Scaling Meta’s Infra with GenAI: Journey to faster and smarter Incident Response
summary
Meta builds a real-time incident onboarding and root-cause pipeline using a fine-tuned Llama 2 7B model plus an internal LM assistant that reads chat, logs and config artifacts and produces a real-time generated summary and follow-up Q&A. Heuristics cut the candidate code-change space from tens of thousands to a few hundred (about a 100x reduction), then iterative LLM re-ranking narrows to ~5 changes. The system surfaces a likely root cause within minutes for 42% of incidents, and nearly half of incidents use the summary with 86% positive feedback.
AI-written summary · about
chapters (6)
- 0:00GenAI in incident response
- 2:13Incident lifecycle challenges
- 4:23Onboarding responders with GenAI
- 9:05Root cause analysis with GenAI
- 16:07Early results and potential
- 19:17Conclusion and next steps
related talks
20mAutomated Incident Management
12mUsing GenAI to Accelerate Privacy Implementations
36mRisks and Opportunities of AI in Incident Management
30mAI and Incident Management: How to Reduce Manual Work and Speed up the Incident...
26mBecoming AI engineers
11mAI-Native Network Operation
23mLive Q&A
19mTeaching AI to Fight Fires: Autonomous Reliability Agents at Meta Scale