Classify, extract, validate, and generate from 150+ financial document types each with its own schema, processing strategy, and review screen. Every value verified before a person ever sees it.
One configurable pipeline from raw package to structured output with a human in the loop wherever you need one.
Package · classified with evidence
Bank statement
pp. 1–12 · 99.1% confidence
Classified
Form 1120
pp. 13–20 · alt: 1120-S (2.1%)
Classified
P&L statement
pp. 21–38 · 98.4% confidence
Classified
Custom doc type
pp. 39–44 · account schema
Classified
CLASSIFY
Know what you're holding.
Every document is processed automatically with a confidence score and ranked alternatives, so routing runs on evidence, not filename guesses.
150+ production document types
Ranked alternative document types, not just one label
Complex, multi-document zip files and PDFs split section by section
EXTRACT
You define what comes out.
Each document type routes through its own strategic processing with structured output enforced against its schema, a rent roll and a credit report are different documents, so they process differently.
A purpose-built pipeline per document type
Multi-model routing with automatic failover
Repeated documents return cached results
Extraction · side-by-side lineage
Source
Fields
Net income
99.2%
EIN
98.7%
Gross receipts
71.4% → review
Verification · per field, per document
Signal confidence
97.4 · auto-accept
Bounding-box traceability
142/142 fields
AI controls re-read
2 corrections logged
Page 7 scan quality
Flagged
VALIDATE
Reading isn't enough.
A silently wrong number costs more than no number. Every field is scored on independent signals, traced to a bounding box in the source, and re-read with AI controls before it moves downstream.
Signal confidence → auto-accept, review, or human review
Click any value, see exactly where it came from
AI controls re-read, correct, and log the history
GENERATE
Shaped to your spec.
Structured JSON enforced to schema on every extraction and the schema is retrievable through the API, so downstream systems can be built before the first document is sent. Formatted outputs and generated documents included.
A doctored pay stub or an edited bank statement is a credit loss waiting to happen. Every file is screened for signs of manipulation as part of processing, not as a paid add-on.
Metadata inspection
Creation and modification history, editing software traces, and timestamps that don't line up.
Font & layout anomalies
Mismatched fonts, misaligned values, and text sitting on top of the original.
Cross-document consistency
The same figure on the bank statement, the tax return, and the application should agree.
Flagged with evidence
Suspected tampering routes to human review with the specific signals attached, never a silent pass.
Pay stub · tamper screen
Employer
Acme Mfg LLC
Net pay
Font mismatch
Pay period
06/01 – 06/15
File metadata
Edited after issue
Page 7 scan quality
Flagged
Routed to human review
evidence attached
Coverage
Over 150 document types.
Template tools break when the document doesn't match the template. Raw LLMs read anything confidently, even when they're wrong. Document AI ships 150+ document types, each with its own schema, processing, and review screen.
Point solutions
Narrow & brittle
Breaks when the doc doesn't match the template
A handful of doc types, verification as an add-on
Dead-ends at a JSON blob no review, no platform
Raw LLMs
Confidently wrong
Well-formatted answers with no way to tell if they're right
No confidence scores, no trace back to the source
Blurry scan in, plausible hallucination out
Uptiq Document AI
Verified & connected
150+ types, each with its own schema & pipeline
Every field scored, traced, and re-read by a judge
Reviewers work in screens designed for each document type — a rent roll renders as a tenant ledger, a bank statement as statement cards. Every field is one click from its source, every comment stays anchored to the value it concerns, and every action lands in the audit log.
Review screens per doc type
Each document type gets an interface designed for it, next to a preview of the original.
Comments on the value itself
Threaded, anchored to the exact field, with resolution tracking.
Ask the document
Query any completed extraction in plain English, grounded in the extracted data.
Role-based access
Full admin down to read-only review, with credentials rotated without disruption.
Audit trail & zero retention
Every action logged. A zero-retention option purges data after processing.
Separated & encrypted
Full tenant separation, encryption at rest, expiring links — dedicated infrastructure when required.
Who it's for
Built for how you'll use it.
POST /v1/extractions · batch: 14 docs
GET /v1/schemas/rent_roll
webhook ✓ delivered · hmac verified · sse:
extracting…
For builders
REST and MCP from day one. Async everything: bulk batches, HMAC-signed webhooks, live progress over SSE. Pull the schema for any document type before you send the first file.
Packages that verify themselves. Your team confirms figures in screens built for the exact document, flags a value with an anchored comment, and asks the file questions in plain English.
Multi-tenant with full separation, dedicated infrastructure when required. Runs on Azure, AWS, or Google Cloud, routes across LLM providers with automatic failover, and keeps spend predictable.
