From Black Box to Glass Box: AI-Driven Observability for Modern APIs
summary
A unified telemetry pipeline is built using OpenTelemetry + Kafka feeding Elastic/Jaeger/Prometheus and single-search Elastic/Elastic stack so on‑call engineers see correlated traces and logs in one place. An ensemble of models (LSTM + Random Forest + Bayesian network) powers early anomaly detection and root‑cause scoring, cutting MTTR from ~180–240 minutes to 45 minutes, weekly pages from ~250 to 50–75, and user‑reported incidents from >50% down to ~5%, while rollout uses six months of shadow‑mode validation.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:11Silent failures in APIs
- 3:20Micro-service challenges
- 6:12Three pillars for observability
- 10:03API reliability and latency
- 11:26AI-driven detection levels
- 13:13Unified data pipeline
- 15:21Implementation and shadow mode
- 17:27Lessons learned and AI's role
- 19:30Future roadmap and predictions
related talks
29mGaining Insights from a Black Box System
30mFrom Black Box To Insight: Observability for AI Agents in Production
13mNext-Gen Observability: reduce MTTR
28mScaling Observability for Real-Time Personalization Engines
42mMachine Learning In Observability Systems: From Insights To Action
01h 01mBetter reliability and uptime with analytics and observability
12mOpen-source observability is the way forward, and here’s why
25mFrom Blind Spots to Total Vision: O11y at Massive Scale