The programme had to make two teams with different cultures deliver as a single pipeline: data and AI for extraction, development for application integration.

I framed the shared architecture, arbitrated practices and aligned responsibilities so that the teams optimised the end-to-end result rather than their own scope alone. The pipeline covers invoice intake, classification, extraction of amounts and references, consistency checks, then loading into a searchable repository.

The design puts business reliability first: a doubtful amount is checked before integration, because wrong data costs more than a document left waiting.