project · 2019-2021
Document-classification ML system
Led development of a production document-classification and entity-extraction system, with reproducible training, evaluation, and deployment workflows.
The problem
Operations teams needed a dependable way to classify incoming documents, extract key entities, and route work to the right queue. The system needed to be accurate, reviewable, and straightforward to retrain as categories changed.
Approach
- Built a text-classification and named-entity-recognition workflow with clear evaluation gates.
- Created reproducible data preparation, training, model-versioning, and promotion steps.
- Worked with a small engineering team to turn model outputs into structured operational inputs.
Outcome
The project established an end-to-end MLOps pattern: measured model quality, traceable releases, and a maintainable path for adding new document types.