At Merqato, I redesigned the Kedro ML architecture, unifying six workflows with reusable features and model lifecycle controls. I built and operated demand and sourcing pipelines that deliver forecasts one to six weeks ahead.
I automated weekly refitting, inference, and forecast publication with Azure Data Factory and Container Instances. I also created an Azure ingestion framework supporting 10 adapters and delivered six integrations across four customers.
For forecast evaluation and monitoring, I combined rolling-origin validation, production replay, and SQLMesh comparisons of served forecasts against actuals across six horizons. I cut SQLMesh forecast-matching runtime from 15–20 minutes to about one minute in development testing.
Previously, I developed revenue forecasting models and data pipelines at FedEx, and integrated AI models from five clients into the Philips platform, delivering seven use cases. My work spans machine learning, data engineering, and Python backend development, with a focus on time-series forecasting and production ML systems.
