AI-powered prior authorization document generator. Generate complete PA packages in 5 minutes instead of 45. Built by physicians, for physicians.
Prior authorization is broken:
- 30-60 minutes per PA case for clinical staff
- 82% of PA denials are overturned on appeal -- but only 11% are appealed (AMA 2023)
- 14.6 hours/week spent by physicians on PA paperwork (AMA Survey)
- 34% of physicians report PA has led to a serious adverse event for a patient
The system punishes patients and providers with paperwork instead of providing care. We're building the fix.
- On-Premise AI -- All models run locally. No PHI leaves your network. Ever.
- Smart Form Builder -- Dynamic forms for 5 procedure categories (imaging, surgery, DME, medications, therapy)
- AI Medical Necessity Letters -- Generated from verified clinical guidelines via RAG, not hallucinated
- Real-Time Code Validation -- ICD-10, CPT/HCPCS, NPI, and RxNorm search and validation
- PDF Document Assembly -- Cover sheet, PA form, justification letter, supporting docs, attestation
- CMS-0057-F Ready -- Built toward Da Vinci PAS (FHIR) compliance from day one
- Select a template -- Choose from 10 pre-built templates across 5 procedure categories
- Fill the smart form -- Dynamic fields adapt based on procedure type and payer requirements
- AI generates justification -- Medical necessity letter backed by RAG-retrieved clinical guidelines with inline citations
- Review and edit -- Rich text editor with tracked changes. Human review is mandatory -- never skippable
- Export PDF package -- Submission-ready PDF/A with all required documents assembled
The AI never invents clinical facts. It only arranges verified facts into a letter.
| Layer | Protection |
|---|---|
| Structured Input | Validated form fields, not free text |
| Retrieval Only | RAG from verified clinical guidelines -- no parametric knowledge |
| Constrained Generation | LLM generates only from retrieved context + patient data |
| Citation Enforcement | Every claim must reference a specific guideline |
| Clinical Validation | Second model validates terminology and logical consistency |
| Code Verification | ICD-10, CPT, NPI codes validated against real registries |
| Human Review | Mandatory review before any document is finalized |
| Layer | Technology |
|---|---|
| Frontend | Next.js 16, React 19, CMS Design System, TailwindCSS 4 |
| Backend | FastAPI, Python 3.12+, SQLAlchemy 2.0, Celery |
| AI Models | On-premise LLMs via vLLM (medical generation + validation) |
| Vector Search | Qdrant (clinical guideline retrieval) |
| Database | PostgreSQL 16 + pgvector |
| Auth | Keycloak 26 (OIDC, RBAC, MFA) |
| WeasyPrint (HTML/CSS to PDF/A) | |
| Design System | CMS Design System (Section 508 / WCAG 2.1 AA) |
| Category | Examples |
|---|---|
| Diagnostic Imaging | MRI, CT, PET, ultrasound |
| Surgical Procedures | Knee replacement, spinal fusion, bariatric |
| Durable Medical Equipment | CPAP, wheelchair, prosthetics |
| Medications | Biologics, specialty drugs, GLP-1 agonists |
| Therapy Services | Physical therapy, occupational therapy, speech |
Built by DevDoc JFMD -- a physician building AI tools to fix broken healthcare workflows.
Early development. The first working pilot targets prior authorization document generation for the 5 procedure categories listed above. This is a sub-project of the broader Doc Assist AI platform.
If you believe in fixing healthcare paperwork with AI, consider sponsoring this project.
Or reach out directly: sponsor@intelmedica.ai
Business Source License 1.1 -- Converts to Apache License 2.0 on March 8, 2030.
Free for non-production and educational use. For commercial licensing, contact hello@intelmedica.ai.
Found a vulnerability? Please report it responsibly. See SECURITY.md.
We welcome contributions. See CONTRIBUTING.md and our Code of Conduct.
Prior Auth Generator by IntelMedica.ai