Schema-first extraction for portfolio reporting.
Nothing rolls up until a human validates it.
Last updated 2026-05-21
What gets extracted
Read more
Canonical operating metrics from quarterly reporting packages, plus company name, period, currency, AI insights, and source quotes per field. Missing values are expected — shown as em dashes in the UI.
End-to-end flow
Read more
Upload PDFs to storage → metadata prep on /batches/{id}/prepare (companies, periods, supporting URLs) → per-PDF context → validation and parsing → reviewers validate each PDF → validated metrics feed portfolio views. See Upload workflow for the three-phase prepare pipeline and statuses.
Related docs
Read more
Product philosophy — why schema-first and high-touch. Upload workflow — intake and statuses. Review workflow — validation rules. Technical docs — data model & Cloud SQL, GCS storage.