Start with the physical system
During my Data Scientist I role, I owned end-to-end delivery for three customer accounts, primarily in Oil & Gas. Onboarding was part of the modeling work: asset hierarchies, signal mapping, and data validation established what the measurements meant.
From historian data to models
I built failure-prediction, anomaly-detection, and equipment-health models over data pipelines integrating Snowflake, Pi Vision, and Canary APIs. These pipelines handled millions of sensor readings per day.
The important connection was between reliable operational data and a model output that an engineer could interpret inside the platform.
Close the loop with the field
I presented results in weekly calls with engineers, site managers, and performance engineers, then iterated on their feedback. The work ran from client onboarding through custom analytics rather than stopping at a model notebook.
A foundation for AI leadership
This experience informed my later work on repeatable analytics packages, canonical onboarding specifications, and agents whose answers need traceable evidence.
Explore the MCP-first delivery workflow →
Illustrated assets and architecture only; no private historian schemas, client data, or production code are shown.