Self-aware alerting for partial discharge & gas density

The brief

Industrial environments generate a firehose of telemetry — gas density readings, partial-discharge measurements, threshold breaches. Traditional rule-based alerting either drowns operators in false positives or misses the patterns that matter. Siemens Energy wanted to know whether agentic AI could change that.

What I built

Agentic and RAG-based AI for safety-critical telemetry

• Designed agentic and RAG-based AI systems for self-aware partial-discharge and gas-density alerting, combining real-time telemetry streams with LLM reasoning.
• Implemented stateful workflows and decision logic enabling AI agents to reason over historical sensor data, thresholds, and alerts — not just react to the latest reading.
• Built scalable ingestion and inference pipelines to transform raw sensor telemetry into actionable alerts using Python and modern LLM orchestration patterns.

Working with domain experts

• Collaborated closely with Siemens domain experts to translate research concepts into deployable, maintainable systems.
• Built new features leveraging telemetry data from sensors and designed a robust alerting system for when thresholds are hit.
• Focused on auditability — operators need to understand why an AI agent raised an alert, not just that it did.