Docgest — the document agent

Docgest turns every inbound document into structured, decision-ready evidence. A hardship letter, paystub, bank statement, or death certificate arrives as pixels; it leaves Docgest classified, its fields extracted, its content screened, its provenance graded — and attached to the case that was waiting for it.

Intake from every direction

Documents reach Docgest however borrowers actually send them:

Source	How it arrives
Borrower upload	The secure upload flow in the self-service portal, or a single-purpose document invitation link sent for exactly what’s missing.
Case workspace	An operator attaches evidence directly to the case or customer.
Email attachment	Arya hands quarantined attachments from inbound email into the same pipeline.

A document invitation is Docgest working proactively: when a workflow needs proof — income for a hardship review, a death certificate for an estate case — the invitation requests that document specifically, and the upload lands already linked to the requesting case.

The pipeline

Every document moves through the same stages, and each stage records its result:

1. Screen — malware and content safety checks run before anything else reads the file; quarantine is the default until they pass.
2. Classify — the document type is identified against the tenant’s catalog: paystub, benefits letter, government ID, certificate, correspondence.
3. Extract — the fields the workflow needs (employer, pay period, net amount, dates, names) are pulled into typed, reviewable values.
4. Verify — provenance is graded explicitly: screened (content checks passed) is never conflated with trusted-source verified (confirmed against an authoritative source).
5. Attach — the analysis lands on the case with its confidence, its concerns, and deep links from every downstream decision that cites it.

Extracted fields become case facts only after human review approves them. Docgest states its confidence and lists its concerns; it does not round uncertainty up to certainty.

Built for the decisions that follow

Docgest’s output is shaped by what consumes it. The hardship workflow reads the evidence analysis check by check — required documents present, fields approved, verified or merely screened, stale or current — and the hardship policy evaluates against those graded facts. Cookie explains the same states to the specialist in plain language: what’s missing, what’s unverified, what needs replacement. Identity verification, privacy requests, and dispute evidence follow the same pattern.

The boundary

Docgest never approves relief, changes account state, or decides a case. Its product is evidence with stated confidence — reviewed under the same AI output review as every agent, suppressible instantly, and cited by ID in the evidence graph wherever a decision relied on it. When an examiner asks “how did you know their income?”, the answer is one link: this document, this extraction, this verification grade, this reviewer.