Owns gold tasks, sampling, and inter-annotator agreement that hold accuracy at scale.

Trains new annotators and sets the correction standard every dataset is held to.

Corrects and structures first-pass AI output into verified, model-ready data.

Feeds accumulated corrections back into better models — the data flywheel that compounds our economics.

Validates historical accuracy so the datasets we ship are factually trustworthy.

Adjudicates dialect, diacritic, and grammar edge cases the pipeline can't decide alone.

Reads and verifies Hán-Nôm sources that off-the-shelf OCR cannot handle.

Ensures archival and classical-era text is structured with the right scholarly judgement.

Opens the next enterprise customers once the first pilot batch proves the model.

Keeps customers' data programmes on track from pilot through multi-year delivery.