KitaDocs
SolutionsBlogBook a demo
AI Underwriter

End-to-end workflow

Intake, extraction, credit picture, memo, export.

  1. Push the borrower, loan terms and any files with POST /intake. Pass external_ref from the LOS or core banking system for idempotent retries.
  2. Poll GET /applications/{id}/documents (or a single document) while files move from awaiting to processing to verified or low_confidence.
  3. Read GET /applications/{id}/credit for deterministic metrics, normalization adjustments and policy routing. Credit math comes from versioned code, not LLM-generated math.
  4. Trigger POST /applications/{id}/memo, then poll GET /applications/{id}/memo/status until synthesis_in_progress is false.
  5. Pull GET /applications/{id}/memo, GET /applications/{id}/conversation, and the export files back into the system of record.

No completeness gate

Processing is continuous. You do not have to collect a full document set before reading a credit picture; every accepted document tightens the assessment as it lands.

Polling guidance

There is no webhook on AI Underwriter today, so extraction and memo synthesis are both poll-based. Poll on an interval rather than in a tight loop, and treat awaiting and processing as the only non-terminal document states.

For memo synthesis, GET /applications/{id}/memo/status is the authoritative signal. synthesis_in_progress going false means the run finished; last_run_status tells you how it finished.

On this page