AI PDF Extractor (Template → Excel)

RCM teams process many document types — patient statements, billing summaries, superbills, authorization files, Cerner reports — into Excel for downstream work. Manual extraction is slow, inconsistent, and hard to govern per user and template.

Built a secured AI PDF Extractor spanning FastAPI document intelligence and App Hub UI/BFF:

• POST /execute — PDF or JSON in, template-specific Excel out (including multi-sheet layouts)
• Mixed pipeline: text extraction, regex layouts, chunked Azure OpenAI prompts, RapidOCR for scans/handwriting, CPT/post-process
• Template assignment is enforced at execute time; manage APIs power the Templates admin page
• Operator UX: drop zone, per-user templates, prompt/schema drawer, Excel preview/download, history; saved_minutes_total validated (no negatives; min 15 when set) and persisted for FTE reporting
• .NET proxy separates ai.pdfextractor.use vs ai.pdfextractor.manage

• Patient statements & billing summaries
• Appointment List / Detailed App List (Medent); HCO Payer & Amount Tracker
• SMDI ledgers; hospital billing with CPT above/below patients
• Practice Fusion scheduler / superbills; Trust, PHNT, NBA superbills; NBA Printed Patient List
• Authorization files (incl. multi-PDF Nutley aggregation)
• Cerner EKG / Medenet-style reports

• Job persistence: status, template/model, rendered prompt, tokens, latency, errors
• Productivity metrics: saved minutes → hours / FTE / man-day style reporting
• Admin job table with prompt inspection for debugging and audit
• Optional Azure Key Vault for sandbox vs production secrets

• FastAPI extraction service and template assignment APIs
• BFF multipart proxy + permission gates for use vs manage
• App Hub PDF-to-Excel, History, and Templates (user assignment) flows
• Telemetry and productivity hooks for operational reporting