Voice
Agent
Production AI for patient booking
Production AI for patient booking
I built and deployed a production Python voice agent for patient booking workflows, focusing on multi-intent conversations, tool-call reliability, memory tracking, and deployment isolation. The system reduced escalation errors, lowered multi-intent latency, improved booking outcomes, and made the agent cheaper to operate at call volume.
Role
Software Engineer Intern
System Surface
Voice AI, tool calls, memory
Infrastructure
GCP and Kubernetes
Delivery Model
Tenant-isolated deployment
The agent became more dependable through schema redesign, prompt iteration, and state-aware memory. Tool-call reliability moved from 70 percent to 85 percent while cost per 1,000 calls dropped by 27 percent. On the product side, escalation errors fell by 60 percent, multi-intent latency came down by 300ms, bookings rose by 22 percent, and no-shows declined by 15 percent.
Study Period
2025 Internship Cycle
Project Status
Rolled out to customers
Primary Focus
Agent reliability
Output Format
Production AI system