ai Shipped · Flagship
AI Incoming Document Processing
OCR, classify, extract structured fields, and auto-route incoming documents to the right claim. Human review only for low-confidence cases.
Key benefits
- · Documents reach the right claim without manual sorting
- · Structured data extraction pre-fills claim fields (dates, amounts, names, case numbers)
- · Confidence-thresholded auto-routing; humans only touch the genuinely ambiguous
- · Crime-type classifier suggests top-3 crime categories on police reports
What it does
Every document uploaded into VCPMS runs through a pipeline that treats the PDF as data, not just an image:
- OCR via Azure Document Intelligence — full-text extraction plus key-value field extraction.
- Classification — identifies the document type (police report, medical bill, prescription, ID, funeral invoice, court order, insurance EOB, etc.) using a custom-trained Azure model when configured, falling back to keyword-and-field heuristics.
- Structured extraction — dates, amounts, names, case numbers, provider IDs — pulled into a JSON payload attached to the document.
- Auto-routing — a DSL-driven routing engine scores each document against live claim/work-item records (fuzzy/Levenshtein match) and auto-links high-confidence matches to the correct claim. Low-confidence documents are queued for staff review with a structured reason.
- Crime-type classification — when a document is classified as a police report, a keyword-pattern classifier (21 crime-type patterns) suggests the top-3 matching crime types with confidence scores, populating the claim workflow.
Who benefits
Program staff who currently spend significant time sorting and associating documents to claims. With AI processing, staff shift from “what is this document?” to reviewing the decisions the pipeline made.
How programs configure it
Per-tenant feature flag App.InDocumentOCR controls OCR; App.Extensions.AiAssistant enables crime-type classification flow. Routing rules live in a tenant-configurable JSON file (in-document-routing-rules.json) so programs can tune matching logic without code changes.