ClearPath

You leave a doctor's office with a diagnosis and zero idea what it'll cost.ClearPath calls your insurance, finds nearby providers, gets real prices, and tells you exactly where to go and what you owe.

Comment

ClearPath

Inspiration

Healthcare has already become agent-to-agent on the institutional side: insurers optimize claims, providers optimize scheduling, and billing systems optimize collections. Patients are still expected to manually navigate calls, benefits, prices, and booking.
We built ClearPath to give patients their own agent — one that can verify coverage, discover provider pricing, and guide booking in a transparent, controlled flow.

What it does

ClearPath is an AI-assisted insurance navigation workflow that:

  • collects insurance + procedure intake from the frontend
  • verifies insurance coverage details through live voice call stages
  • runs provider pricing calls and compares options
  • estimates patient out-of-pocket cost
  • ranks providers and highlights the recommended option
  • supports booking with a manual trigger to avoid call overlap during critical moments

The current demo includes live telephony bridging so calls can happen in real time.

How we built it

  • Frontend: Next.js + React + TypeScript
  • Backend orchestration: modular adapters + orchestrators for insurance, pricing, and booking
  • Voice + telephony: Twilio Media Streams bridged to xAI Grok Voice via a custom bridge service
  • Stage design: insurance verification, provider pricing, and booking stages with timeline status in UI
  • Reliability controls: fallback parsing, stage-by-stage metadata, manual booking-call trigger

Out-of-pocket estimation follows:

[ R = D_{total} - D_{met} ]

[ \text{if } P \le R,\quad C = P ]

[ \text{else } C = R + (P-R)\cdot \frac{k}{100} ]

Where:

  • (P) = procedure price
  • (R) = remaining deductible
  • (k) = coinsurance percentage
  • (C) = estimated patient cost

Challenges we ran into

  • Real-time call orchestration across Twilio + voice model timing
  • Handling non-perfect structured outputs from voice interactions
  • Avoiding overlapping calls during demo pacing
  • Tunnel/runtime friction (URL changes, process restarts, port conflicts)
  • Balancing realism with deterministic demo reliability

Accomplishments that we're proud of

  • End-to-end staged call workflow running in a live demo setup
  • UI Live Call Timeline that makes backend orchestration visible
  • Clear patient control point: booking call is explicitly manual trigger
  • Strong modular architecture for fast iteration
  • Fully working fallback path when upstream responses are noisy

What we learned

  • Voice systems need explicit stage prompts and strict extraction contracts
  • Demo UX clarity matters as much as model capability
  • A patient-facing agent must reduce friction, not add complexity
  • Manual control points (like booking trigger) are essential for trust and pacing

What's next for ClearPath

  • Replace mock provider discovery with real Maps/Places integration
  • Improve structured extraction quality and validation
  • Add persistent call logs/transcripts for auditability
  • Expand to prior auth and broader procedure categories
  • Productionize observability, retries, and security posture

Built With

Share this project:

Updates