Modern production systems generate massive volumes of telemetry, but most monitoring tools still rely on static thresholds and manual rules. This lightning talk demonstrates how lightweight AI heuristics can be used to detect abnormal runtime behaviour in applications and services without complex data science pipelines.

Using practical examples, I’ll show how simple ML models and heuristic signals—such as entropy shifts, API call frequency changes, and execution pattern anomalies—can be applied to runtime logs, process metrics, and telemetry streams to identify suspicious or malicious behaviour early.

The focus is on engineering-friendly implementation: how to deploy these techniques alongside existing monitoring stacks, trigger intelligent alerts, and reduce noise using AI-assisted signal scoring. This talk is aimed at software engineers looking to enhance observability and security using practical, production-ready AI techniques.