Every year in October, JAX London is the go-to event for Java and Software Architecture enthusiasts! It is a four-day conference for cutting-edge software engineers and enterprise-level professionals.
This year, at JAX London’s 2026 conference, our architect Ram Lakshmanan was invited to talk on the topic: “LET’S FORECAST PRODUCTION OUTAGES USING MICRO-METRICS, PATTERNS & ML”.
Production outages rarely happen without warning — the signals are hidden in plain sight across metrics, logs, and system behaviour. In this session, you will learn comprehensive, real-world approach to forecasting production outages using machine learning and AI techniques applied across multiple data sources. We will cover how anomaly detection on system metrics surfaces early warning signs, how Levenshtein-based clustering and scoring uncovers abnormal patterns buried in high-volume application logs, and how linear regression on garbage collection logs reveals memory pressure trends before they become failures. You will learn how these models work together as a layered early-warning system and how you can implement them incrementally in real-world JVM environments without building a heavy data science platform.
Here is the deck shared in the conference: