From raw data to a decision.
Pipelines that do not silently break, models that survive contact with production, and an honest account of what the numbers can and cannot support.
How I work
Most of this job is not the model.
Pipelines first
A model is only as good as the data arriving on time and intact.
Data quality
Validation, drift detection and alerting before anything is trained.
Right-sized models
A gradient-boosted tree that ships beats a transformer that does not.
Measured properly
Offline metrics that actually predict the online result.
Reproducible
Versioned data, code and parameters. Any result can be rebuilt.
Honest about limits
Where the model is uncertain, and what it should not be used for.
What colleagues say
From the people on the other side of the handover.
“Told us our problem did not need machine learning. Saved us two quarters.”
“The only pipeline in the company that has never woken anyone up.”
“Documents assumptions properly. Rare and enormously valuable.”
Talking about a data problem
Models, pipelines and the stack behind them are on the portfolio.
See every project