AI-powered Prior Authorization, in seconds.

Prior authorization is one of the biggest bottlenecks in modern healthcare — a process that takes 7–14 days, buries clinical staff in paperwork, and delays patient care. Jolt eliminates that friction by combining patient data, clinical documents, and insurance policy requirements into a single AI-driven decision in under 30 seconds.

Built at Hacklytics 2026 — Georgia Tech's annual data science hackathon.

https://jolt-gray.vercel.app/

Patient Search → Coverage Verification → Procedure Selection → AI Evaluation → Decision Report

1. Search — Look up a patient by name with real-time autocomplete. Patients are sorted by most recent evaluation so active cases surface first.
2. Profile — See demographics, active insurance coverage, and all uploaded clinical documents in one view.
3. Select Procedure — Pick a CPT code (with optional laterality) from a searchable catalog.
4. Evaluate — Jolt's RAG pipeline cross-references the patient's clinical records against the payer's policy requirements using Claude AI.
5. Decision — Get a structured report: Likely Approved / Uncertain / Likely Denied, an approval probability score, supporting evidence, and a list of missing information — each with a direct upload button.

• Real-time patient search — Autocomplete by first or last name, sorted by most recent evaluation
• Coverage verification — Active insurance plans, payer name, plan type, and member ID at a glance
• Clinical document viewer — Browse and read uploaded medical records linked to each patient
• Evaluation history — See past prior auth runs per patient with outcomes and timestamps

• RAG pipeline — Policy documents and clinical records are chunked, embedded (VoyageAI), and stored in Supabase with pgvector for semantic retrieval
• Claude-powered analysis — Each payer policy criterion is evaluated in parallel against the patient's evidence
• Evidence citations — Every reason links back to the exact source quote from clinical records or policy documents
• Approval probability score — A 0–100% confidence score alongside the qualitative verdict
• Missing info detection — Surfaces exactly what documentation is still needed, with inline upload buttons to resolve each item

• Clinical document upload — Drag-and-drop PDF or image upload, chunked and embedded automatically
• Policy document upload — Upload payer-specific policy PDFs indexed by CPT code
• Letter of Medical Necessity — Generate and download a pre-filled PDF letter for any completed evaluation

• Full audit trail — Every prior auth run is stored with patient, payer, CPT, timestamps, and outcome
• Filter & search — Filter history by status, recommendation, or patient name
• Drill-down view — Click any historical run to see the full decision report

Jolt/
├── backend/
│   └── src/
│       ├── app.ts                    # Express server entry point
│       ├── db/
│       │   └── supabase.ts           # Supabase client
│       ├── routes/
│       │   ├── patients.ts           # Patient search, profile & document endpoints
│       │   ├── evaluate.ts           # Prior auth evaluation & polling endpoints
│       │   ├── cpt.ts                # CPT code search
│       │   └── upload.ts             # Clinical & policy document upload
│       └── services/
│           └── ragPipeline.ts        # RAG evaluation pipeline (chunking, embedding, retrieval)
├── frontend/
│   └── src/
│       ├── components/
│       │   ├── EvaluateResults.tsx   # Decision report with evidence, missing info & upload sheet
│       │   ├── CptSearch.tsx         # Procedure search with laterality toggles
│       │   ├── AnatomyHighlighter.tsx # Body diagram for procedure visualization
│       │   └── ui/                   # shadcn/ui component library
│       ├── pages/
│       │   ├── LandingPage.tsx       # Marketing / entry page
│       │   ├── MainPage.tsx          # Patient browser & evaluation launcher
│       │   ├── EvaluationResultPage.tsx # Decision report + evaluation history
│       │   └── AdminPage.tsx         # Document upload management
│       └── lib/
│           ├── api.ts                # Typed API client (with mock mode)
│           ├── types.ts              # Shared TypeScript interfaces
│           ├── reasonCategories.ts   # Reason categorization logic
│           └── procedureDescriptions.ts # CPT code descriptions
├── docs/                             # PRD, Tech Design, Research
├── agent_docs/                       # Detailed reference docs per feature area
├── AGENTS.md                         # Build phase tracker
└── CLAUDE.md                         # Claude Code configuration

• Node.js 18+
• A Supabase project with patients, coverage, document_chunks, and prior_auth_requests tables
• API keys for Anthropic, VoyageAI, and OpenAI

cd backend
npm install

Create a .env file:

PORT=3001
SUPABASE_URL=your_supabase_url
SUPABASE_SERVICE_KEY=your_supabase_service_key
ANTHROPIC_API_KEY=your_anthropic_key
OPENAI_API_KEY=your_openai_key
VOYAGE_API_KEY=your_voyage_key

Start the dev server:

npm run dev

Backend runs at http://localhost:3001.

cd frontend
npm install

Create a .env file:

VITE_API_URL=http://localhost:3001

Start the dev server:

npm run dev

Frontend runs at http://localhost:5173 (or the next available port).

Built in 36 hours at Hacklytics 2026 — Georgia Tech's annual data science hackathon.