RPM's team no longer re-keys workers' compensation policy details by hand. Upload a PDF, and the platform extracts policy headers, multi-state classifications, premium totals, and exclusion schedules — with every value linked back to its exact location in the document.
Document viewer · Extracted information
Document viewer
Extracted information
Data extractedPolicy No.
WC-2024-8841
Named Insured
Summit Logistics LLC
FEIN
84-2910337
Classifications
10+
Structured fields extracted per policy document
100%
Extracted values linked to source locations in the PDF
Multi-state
Classification and premium tables across states
1 upload
From PDF to structured, review-ready output
Workers' compensation policy documents arrive as long, layout-heavy PDFs. Each one carries critical data — policy number, coverage period, FEIN, named insured, state-by-state classification codes, rate tables, premium totals, and officer exclusion schedules. For RPM, reviewing and transferring this information was a slow, error-prone process.
Operations staff spent significant time reading multi-page policies and manually entering data into downstream systems. WC PDFs are not simple forms: tables span pages, descriptions continue across rows, the same code can appear in different states, and premium sections mix summary lines with adjustments that must not be confused with totals. RPM needed extraction that understood insurance document structure — and let reviewers verify every value against the original source.
01
PyMuPDF extracts full document text and word-level coordinates. OpenAI structured output — Pydantic-validated — turns unstructured policy text into typed fields: headers, classification rows, premium tables, and exclusion schedules.
02
Every field stores both a human-readable value and the exact PDF text. The UI stays clean while source matching stays precise for highlights and audit.
03
Extracted values are matched back to word coordinates. Reviewers click any field, classification row, or premium cell and the platform jumps to the correct page and highlights the source.
04
The engine encodes domain logic: multi-row classification merging, state-level deduplication, filtering non-classification rows, and strict rules for Total Class Premium vs. adjustment lines.
01
A drag-and-drop zone accepts policy PDFs. After authentication, users click Extract and the backend processes the document. Extracted data appears in a structured panel alongside the original PDF — no tab-switching or re-uploading.
Upload · WC policy PDF
Drop workers' comp policy PDF
policy-summit-logistics.pdf · 2.4 MB
02
The document viewer renders the PDF with page navigation, a data-extracted status badge, and a toggle for highlights. Reviewers work in a split view: document on one side, structured output on the other — policy headers, premium tables, classifications, and exclusion schedules.
Document viewer · Extracted information
Document viewer
Extracted information
Data extractedPolicy No.
WC-2024-8841
Named Insured
Summit Logistics LLC
FEIN
84-2910337
Classifications
03
Clicking any policy field, classification row, or premium value selects the corresponding data, navigates to the correct PDF page, and highlights the source region. Reviewers confirm AI output against the original document — not a black-box result.
Source-linked verification
PDF · highlighted source
Click a field to jump here — page 4, classification row
Selected · Code 8810 · CA
Matched in document
8810 — Clerical office employees · California · $1.42 per $100
Display value
$1.42
Raw PDF text
1.42