Back to projects

Medical Practice Voice Agent

An AI front desk for a private medical practice. It checks a caller's insurance, estimates what they'll owe, and books the visit.

Try the demo

Overview

An AI front-desk agent that handles a patient's entire pre-visit intake in one call: it answers questions, verifies insurance, estimates the visit's cost, and books the appointment. When it's unsure, it hands off to staff instead of guessing.

The Problem

The front desk is where a practice's revenue cycle starts, and where much of it breaks. Registration errors surface weeks later as denied claims: 41% of providers reported denial rates of 10% or higher, and bad registration data drove nearly a third of denials (Experian Health, 2025). Checking eligibility by hand also costs about $6.78, versus $0.34 electronically (CAQH Index).

Patients on high-deductible plans want to know what a visit will cost, but a busy front desk usually can't explain their benefits over the phone. The burden falls hardest on independent practices: only 42% of physicians were in private practice in 2024, down from 60% in 2012 (AMA).

Why I Built This

Investors and top healthcare AI startups are betting that voice AI can now handle this conversation, and that the value lies in connecting it to payer and scheduling logic. Assort Health ($1.2B valuation) and EliseAI ($4B) automate the patient journey for specialty practices, from scheduling to follow-up. Prosper AI, backed by a16z, also calls insurers to verify benefits, chase prior authorizations, and follow up on claims. Sierra ($15.8B) is a general-purpose enterprise platform serving health plans, providers, and revenue cycle companies. Epic is building agents into MyChart.

I built this to learn the workflow end to end and see firsthand where a voice agent works, and where it falls short, as healthcare's front door.

Approach

  • Voice and Text Agent (ElevenLabs + Claude): Asks one question at a time, reads details back to confirm them, and explains outcomes in plain language.
  • Insurance Rules Engine (Mock Clearinghouse and Fee Schedule): Parses the eligibility response to determine active coverage, copays, remaining deductible, and coordination of benefits when a patient has two plans. It then produces a visit cost estimate and flags cases that need staff review.
  • Booking and Handoff (Cal.com + Supabase): Books or holds appointments and turns anything uncertain into a staff case so nothing is dropped. Patients can also self-schedule.
  • Evaluation Harness: Simulated patients call the agent, some deliberately giving wrong or incomplete information. Every call is graded on safety, accuracy, and rule adherence.

Stack: ElevenLabs, Claude, Vercel, Supabase, Cal.com, Claude Design