Building a 24/7 AI Voice Receptionist for Women's Health Clinics
From problem to live product — the full story of HerCare Voice.
The Problem
Unanswered Calls in Women's Health Clinics
Women's health clinics are busy environments. Patients call to book appointments — but if the call goes unanswered after hours or during peak times, the patient moves on and calls a competitor.
Missed Revenue
Every unanswered patient call is a lost booking. Patient leakage directly impacts clinic revenue and retention.
No Structured Data
Unstructured voicemails or hand-written notes get lost or forgotten, creating gaps in follow-up booking logs.
No 24/7 Coverage
Human receptionists only work office hours, leaving the clinic unreachable for evening and weekend patient inquiries.
The Solution
Introducing Maya — The Voice AI Coordinator
We built HerCare Voice — an AI voice agent named Maya that answers every inbound call 24/7. Maya talks to the patient in natural conversation, checks real availability, books the appointment, and logs everything automatically with zero human involvement.
Key product decision: We explicitly chose voice AI over a chatbot or web form because patients calling a women's health clinic expect warmth, empathy, and reassurance. Voice reduces booking friction and feels more appropriate for sensitive healthcare interactions.
How It Was Built
Step-by-Step Production Timeline
Agent Provisioning
Created the voice agent in Retell AI with GPT-4.1 Mini and Cartesia Hailey voice.
Persona Prompting
Wrote Maya's system prompt defining her persona, tone, and what patient details to collect.
Phone Setup
Provisioned a dedicated US phone number through Retell AI ($2/month) to receive live calls.
Flow Design
Built the call flow node by node in Retell AI's visual canvas structure.
Scheduler Sync
Connected Cal.com APIs for real-time availability checking and automated booking.
Data Extraction
Configured Post-Call Data Extraction for 7 custom patient demographic and scheduling fields.
Backend Pipeline
Built the n8n webhook workflow to log every call data segment automatically to Google Sheets.
End-to-End Testing
Conducted live calls confirming flow booking and database logging worked perfectly.
Landing Deployment
Built and deployed the landing page and dialer simulator to hercare.buildsmartai.app.
Key Technical Challenges
Overcoming Latency and Function Edge Cases
Cal.com Node Issues
Challenge: Cal.com was not firing during calls when configured as background LLM tool descriptions.
Solution: We wired Cal.com functions as explicit nodes in the canvas. Once added as dedicated nodes, availability checks worked perfectly.
Function Collisions
Challenge: Structured Output configurations were breaking external Cal.com API call functions.
Solution: Retell AI's Structured Output toggle must be turned OFF for external API functions to fire and book successfully.
Webhook Data Gaps
Challenge: Using the wrong webhook event meant no patient data was being sent to Google Sheets.
Solution: Switched from call_ended (which fires on disconnect without payload) to call_analyzed which carries all extracted variables.
Voice Latency
Challenge: Default Retell AI synthetic voices caused audio buffering and choppy speech latency in live calls.
Solution: Switched to Cartesia Hailey voice. This optimized model-to-voice stream and eliminated buffering completely.
Results
System Performance and Outcome Metrics
What I Would Build Next
Future Roadmap & Scaling Features
- SMS confirmation: Auto-send text confirmation and map details to the patient right after booking.
- Clinic dashboard: Real-time operations web application showing call volume, analytics, and booking rate.
- Edge-case handling: Support for partial names, accents, and request redirections.
- HIPAA compliance review: Formally audit data storage parameters before live clinical deployment.