Keynote: AI Enabled Observability Explainers - We Actually Did Something With AI!
Talkat KubeCon + CloudNativeCon Europe 2025 · on Wednesday, 2025-04-02· at 09:33· Level 0 | ICC Auditorium·
summary
eBay builds deterministically reliable observability “explainers” that combine algorithms and LLMs: trace explainer prunes traces via a critical path algorithm (handles 3,000–8,000-span traces) and dictionary-encodes services, then uses LLMs only for summarization. Other building blocks: log, metric and change explainers, plus compound dashboard explainers and alerts with embedded faulty traces. Prior ML system Groot cut detect time under four minutes; a 1,300-span DB slowdown is resolved by the explainers locating the slow service and a timeout in logs.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:46Observability at eBay
- 3:30ML to LLMs in observability
- 6:22Deterministic AI explainers
- 8:28AI and engineering for explainers
- 10:58Layering explainers in triage
- 12:26Future possibilities
- 14:11Path forward
related talks
28mWhere are we going with observability?
39mKeynote 3: How did you know? Explaining Black Box Model Predictions in Python
33mWhere Are We Going With Observability?
18mAI Observability: How To Fix Issues With Your ML Model
06mKeynote: Beyond Observability: Intelligent Progressive...
21mAI-Powered O11y with LLMs: Resilience in Cloud PaaS Archs
29mElevating generative AI and LLM observability
30mAI Agents for Incident Management