Document intelligence pipeline (Aurora)

Problem. A healthcare services organization received a high volume of inbound clinical documents as unstructured faxes. Routing and keying them by hand was slow and error-prone.

Approach. Designed and shipped a production document-intelligence pipeline: custom classification models sort each inbound document by type, purpose-built extractors pull the fields that matter, and the results land in a relational store the downstream apps read. Built on Azure - Document Intelligence for the models, Functions for orchestration, Blob Storage for documents, PostgreSQL for structured output, Key Vault for secrets - with separate development and production environments and infrastructure defined as code.

Impact. Owned the migration from development to production - data, trained models and document storage - and the incident response that followed, including a same-day fix for a classification outage caused by a model-configuration mismatch between environments. Established the rule that models ship to an environment before the configuration points at them.