SmartClaim AI

Medical billing compliance is one of healthcare's most expensive failure modes. Clinicians and billing teams must manually verify that each claim satisfies CMS Medicare LCD requirements — a time-intensive process requiring familiarity with dense regulatory documents like the Novitas LCD for Cellular and Tissue-Based Products (Skin Substitutes). Manual review introduces inconsistency, omission errors, and audit exposure. A single non-compliant claim can trigger denial, appeal cycles, and regulatory scrutiny, while the aggregate cost of compliance failures across a billing operation runs into millions annually.

Built a full-stack AI compliance platform on Next.js 16 with TypeScript, using Vercel AI SDK for streaming OpenAI responses that give users real-time feedback as the analysis progresses rather than waiting for a full response: **Compliance Engine:** Clinical notes and billing checklists are ingested and analyzed against CMS Medicare LCD Novitas guidelines in real time. The AI model flags non-compliant elements with detailed clinical reasoning and highlights the precise text requiring attention. **Authentication:** Full auth flow implemented with NextAuth v5 (Auth.js beta) — sign-up/sign-in with bcryptjs password hashing and secure session management. No patient data exposed without authentication. **Data Layer:** Prisma ORM with SQLite for local development and Supabase PostgreSQL in production, enabling schema migrations and type-safe database access throughout. **Form & Validation:** react-hook-form + Zod schema validation ensures clinical inputs are validated before submission, preventing malformed queries reaching the AI layer. **Markdown rendering:** AI responses use react-markdown with remark-gfm for structured, readable compliance reports with proper heading hierarchy and bullet lists.

A live compliance platform deployed at smartclaimai.vercel.app that automates the manual claim review process against CMS Medicare LCD standards. The streaming AI response gives billing teams real-time feedback during analysis, and precise text highlighting eliminates the ambiguity of generic "non-compliant" flags. Full authentication and Supabase PostgreSQL make the platform production-ready for clinical deployment.