Projects · Agentic AI · In progress

Industrial maintenance agent

An agent that ranks a turbofan fleet by wear, explains each engine's health from sensor data and cites the maintenance manual.

Python · Anthropic SDK · RAG · NASA C-MAPSS

Problem

Maintenance teams need to know which engines to inspect first, and why. Raw remaining-useful-life predictions are hard to act on: they don’t explain themselves and don’t connect to the procedures an engineer actually follows.

Data

NASA’s C-MAPSS turbofan degradation dataset: multivariate sensor time series for fleets of simulated engines run to failure, plus a maintenance manual used as the retrieval corpus.

Approach

  • An agentic system built directly on the Anthropic Python SDK, with three custom tools:
    • Fleet ranking — orders active engines by current wear.
    • Health assessment — summarises an individual engine’s condition.
    • Sensor readings — retrieves the underlying measurements for inspection.
  • A 0–100 wear index for every active engine, calibrated on engines that went on to fail, so the score reflects expected health before deterioration sets in.
  • RAG retrieval over the maintenance manual, so recommendations are grounded in documented procedures.

Results

Results write-up coming soon. A live, embedded demo is planned.

What I’d do next

  • Embed the live agent demo on this page.