Inspiration

My aunt lost her house when she had breast cancer because couldn't afford the bills. She went into debt because insurance wouldn't pay.

81.7% of appealed Medicare Advantage denials get overturned but only 11.7% are ever appealed

Doctors and staff spend over 13 hours a week dealing with claim denials.

What it does

it reads claims/denial letters and compares them to our 120,000+ claim database, uses grok to create an appeal.

word coach to improve likelihood of approval

grok voice will even call the insurance provider

How we built it

  • Stack: React + Vite frontend on a Skateboard shell, Hono + SQLite backend
  • Intelligence: xAI Grok structures the analysis, plain-English explanation, Word Coach, and draft appeal from a curated RAG knowledge base of cancer denial patterns
  • Voice: Grok Voice over a realtime WebSocket — server-minted ephemeral tokens, mic PCM16 streamed with server-side VAD and audio playback
  • Demo data: A full Maria Santos packet (denial letter, prior auth, EOB, CMS-1500, weak vs. strong appeal drafts) plus real CMS-style PDFs, seeded on load; optional EMR import feeds medical records into the Grok analysis

Challenges we ran into

  • Making Grok specific, not generic: the whole point is which words won — we had to inject appeal-strong patterns and contrastive coaching rather than letting it default to a polite-letter template.
  • Domain accuracy: Medicare vs. Medicare Advantage vs. commercial plans handle prior auth completely differently.

Accomplishments that we're proud of

  • A working Word Coach that shows weak → winning language, not just a generated letter
  • Grok Voice that's a real accessibility feature, not a gimmick built for patients too overwhelmed to read dense letters
  • An end-to-end demo: documents → one-click analyze → Analysis drawer → voice Q&A → copy-ready appeal
  • Grounding every claim in real data (KFF/CMS, AMA, ProPublica's EviCore investigation) ## What we learned
  • The 81.7%/11.7% gap is a UX problem, not a legal one: people don't appeal because the process is overwhelming, so voice + one-click matter more than legal sophistication.
  • its not the denial it's the delay. people often die because of the insurance approval delays

What's next for claim decline assistant

  • Auto-fax/portal submission so the appeal actually files itself, not just drafts.
  • Deadline tracking + reminders (appeals have hard windows that patients miss).
  • EMR/payer API integrations (FHIR) to pull records automatically instead of manual import.
  • Expand the knowledge base beyond cancer to all high-denial specialties.
  • Outcome tracking loop — feed real approve/deny results back to sharpen the Word Coach.
  • HIPAA compliance + provider/patient-advocate pilot.

Built With

Share this project:

Updates