Auth Pilot Health

Agentic prior authorization case management that keeps a clinician in charge of every denial and appeal.

Auth Pilot Health is a provider-side prior authorization workspace. We treat every request as a living case that moves through intake, evidence gathering, payer-rule matching, submission, decision, and appeal. Agents and robots do the paperwork; a clinician stays legally and visibly in charge of every medical-necessity decision through a mandatory human gate.

Built for UiPath AgentHack (Track 1: Maestro Case).

Inspiration

We kept hearing the same story from clinic staff: prior authorization eats hours, denials pile up, and fewer than 12 percent of denials ever get appealed because nobody has time. Then 2025 laws started banning fully automated medical-necessity denials and CMS announced its WISeR AI prior-auth pilot for January 2026. We wanted to build the provider-side answer: automation that does the paperwork but legally keeps a clinician in charge of every decision.

What it does

Auth Pilot Health runs every prior authorization as a living case that moves through intake, evidence gathering, rule matching, submission, decision, and appeal. A clinical-evidence agent extracts diagnoses and documentation from the chart, an external payer-rules agent scores the case against the insurer's medical-necessity policy and flags exactly what is missing, and RPA robots submit and poll the portal. When a denial comes back, an agent drafts a citation-backed appeal, but the case pauses at a human task so a clinician must review and sign before anything is sent. Coordinators and utilization-review nurses get faster clean submissions and rescue winnable denials.

How we built it

We built the orchestration around UiPath Maestro Case as the case lifecycle and human-task spine, with an Agent Builder agent for evidence extraction, an external CrewAI agent for payer-rules matching under UiPath governance, RPA robots for the portal work, and Claude on the AI Trust Layer for reasoning. The app itself is a Next.js 14 and TypeScript front end with Tailwind and shadcn-style components, a Zustand case store that models the full lifecycle and audit trail, and server route handlers that call Claude live when you add a key or fall back to deterministic synthetic results. Cursor scaffolded and shipped the whole thing through UiPath for Coding Agents.

Challenges we ran into

The hardest part was making the human gate feel native rather than bolted on: the case has to genuinely pause, surface the right approval or sign-off task, and refuse to send anything until a clinician acts, all while staying auditable. We also worked to blend a low-code Agent Builder agent with an external CrewAI agent under one governance layer so the matching step stayed a single, well-scoped responsibility instead of sprawling.

Accomplishments that we're proud of

We are proud that the human-gate compliance story is the centerpiece, not an afterthought, and that you can walk a case end to end in under a minute: drop a referral, watch the agents extract and flag a missing document, approve at the gate, take a denial, draft an appeal, and sign off, with every step timestamped in the audit trail. The whole thing runs out of the box on synthetic data and degrades gracefully without any API key.

What we learned

We learned that durable, long-running case orchestration is a different discipline from one-shot LLM calls: state, interruptions, pause and resume, and audit are most of the work. We also learned how much trust comes from making the human decision point explicit and legible rather than hiding the automation behind a single approve button.

What's next for Auth Pilot Health

Next we want to:

1. Wire the live Maestro Case instance APIs and Orchestrator queues to a real Labs tenant.
2. Connect a FHIR adapter so evidence extraction reads real (de-identified) charts.
3. Expand the payer-policy library and let the CrewAI agent cite the exact policy version.
4. Add peer-to-peer review scheduling for denials that need a physician call.
5. Ship the denial-pattern dashboard into a root-cause workflow that proposes template fixes.

Built with

Next.js 14, TypeScript, Tailwind CSS, shadcn-style components on Radix UI, Zustand, UiPath Maestro Case, UiPath Agent Builder, UiPath Orchestrator RPA, UiPath for Coding Agents, CrewAI, Anthropic Claude, and Cursor.

Built With

• anthropic-claude
• crewai
• cursor
• next.js-14
• radix-ui
• shadcn-ui
• tailwindcss
• typescript
• uipath-agent-builder
• uipath-for-coding-agents
• uipath-maestro-case
• uipath-orchestrator-rpa
• zustand