Business Overview

Our client is a premier US healthcare technology firm providing Hub Services—a centralized support model that coordinates the complex logistics of specialty medication access. Supporting over 80 major drug brands, they act as a critical intermediary between manufacturers, healthcare providers, and insurance payers, helping navigate administrative complexity. In the high-stakes specialty pharmaceuticals sector, the company manages the intricate onboarding process, including patient registration, benefit verification, and prior authorization.

Previously, the client relied on manual intervention when data could not be retrieved via APIs or third-party sources, resulting in delays that directly affected time to treatment initiation (TTI). To accelerate access to life-critical medications, the client aimed to replace manual outreach with an AI-powered voice agent and turned to NIX, their trusted long-term technology partner, to bring this solution to life.

Project Scope

This greenfield initiative focused on developing an autonomous AI voice agent for healthacre. Having demonstrated our deep expertise in conversational AI and SIP telephony, we integrated directly into the client’s internal engineering team to deliver the following:

  • 1

    Architectural design and implementation: Contributed to the agent’s end-to-end architecture, including a compliant voice generation service and a comprehensive AI observability framework.

  • 2

    Infrastructure and telephony: Managed core infrastructure and SIP telephony integration to ensure seamless communication flows.

  • 3

    Dialogue and evaluation: Fine-tuned complex dialogue behaviors and developed rigorous evaluation harnesses to ensure data accuracy and conversational reliability.

Challenge

  • Pioneering industrial-grade voice AI: With no established industry playbook for autonomous voice agents, we designed the architectural standards for reliability from the ground up.
  • The quality of voice: To prevent insurance operators from hanging up, the agent’s voice had to maintain a convincingly human, professional tone over standard phone lines.
  • Telephony-induced complexity: We had to ensure the AI remained focused despite lag and background noise while navigating multi-layered interactive voice response (IVR) menus.
  • Dynamic conversational logic: The system had to master context management, turn-taking, and interruption handling.
  • Strict security compliance: To satisfy protected health information (PHI) regulations, the client required all components to be containerized and self-hosted within their private AWS environment.
picture 02

Solution

We led high-impact workstreams across infrastructure, security, and conversational design. Our approach centered on building a robust, pluggable architecture that allowed the voice agent to interact with various AI models and telephony systems. This modular design enables upgrading individual components—such as switching to a faster language model—without rebuilding the entire system.

High-performance and Compliant Infrastructure

picture 03

Data privacy is paramount in healthcare. To ensure 100% compliance with BAA and HIPAA standards, we engineered a self-hosted text-to-speech (TTS) service.

  • In-house processing: We designed and deployed the voice generation system directly within the client’s private AWS environment. This ensures that sensitive patient data never leaves their secure cloud, meeting all regulatory requirements.
  • Professional voice quality: We fine-tuned the synthesized voice to sound professional and empathetic, resulting in higher engagement rates when speaking with insurance operators.
  • Scalable deployment: By using Terraform for modern infrastructure-as-code, we replaced manual work with a set of written configuration files, making the system easy to manage and automatically scale as call volumes increase.

Intelligent Dialogue and Behavioral Tuning

The sophisticated voice agent’s task isn’t limited to speaking—it must also listen, understand context, and master the natural flow of conversation required to navigate a live phone call.

  • Agent behavior: We meticulously tuned the AI’s instructions to manage interruptions and navigate complex IVR menus.
  • Natural turn-taking: The agent was trained to recognize when to wait for an operator’s answer and when to politely persist to get the required clinical data.
picture 04

Advanced Observability and Evaluation

picture 05

Testing a voice AI is far more complex than testing standard software. We introduced a dedicated evaluation layer using a platform called Langfuse, which has since become a standard for LLM engineering, observability, and automated quality evaluation across the client’s company.

  • AI-driven grading: Our experts used an LLM-as-judge framework to grade the voice agent’s responses, ensuring accuracy and professionalism in every turn of the conversation.
  • Simulation testing: We built a simulation harness to replay and test many dialogue scenarios, ensuring the agent remains consistent even when insurance operators provide unexpected or difficult answers.
  • Full transparency: Every call is traced, recorded, and stored securely in AWS S3 for auditing and QA.

Seamless Ecosystem Integration

The voice agent is a critical link in a larger automation chain, functioning as one stage in a high-speed data pipeline, ensuring a frictionless flow of information across a sophisticated tech stack:

  • Intelligent handoff: The process is triggered by an upstream AI that parses enrollment forms. If data is missing, the system initiates a smart handoff to the voice agent for completion.
  • Cognitive core and telephony: Powered by AWS Bedrock (Claude) for complex medical reasoning, the system uses LiveKit for real-time audio and Twilio for secure outbound calling.
  • Speech and voice services: We integrated Deepgram and AssemblyAI for high-accuracy hearing (speech-to-text), paired with ElevenLabs and our self-hosted Qwen3 service for clear, secure voice generation.
  • Data delivery and storage: Verified data is published instantly to internal systems via Kafka (a high-speed data streaming platform), while full recordings and audit logs are archived in AWS S3 under strict HIPAA retention policies.
  • Observability: Every interaction is monitored via Langfuse, providing a real-time dashboard for tracking AI accuracy, system health, and call performance.
picture 06

Outcome

Together with the client’s team, NIX delivered a fully functional voice AI in healthcare that autonomously handles multi-turn insurance calls, converting complex conversations into structured, auditable clinical data with secure recordings and full observability. The solution reduces the need for manual verification teams, lowers operational costs, and helps accelerate treatment initiation—the client’s key KPI. It is currently in the active demo phase and undergoing final stakeholder refinements.

Major Achievements

90%

Faster processing time

70%

Automation rate

95%+

Reduction in per-call cost

30–40%

Total cost savings

Team:

Team:

1 expert ( Account Tech Lead )
Tech stack:

Tech stack:

Python, LiveKit Agents Framework, AWS Bedrock, Nova Sonic, Langfuse, FastAPI, Node.js, Terraform, AWS ECS, TypeScript, Tailwind

REQUEST A CONSULTATION

Contact us   

Relevant Case Studies

View all case studies

Pharmaceutical Software Modernization and AI Implementation

Healthcare

Success Story Pharmaceutical Software Modernization and AI Implementation image

Population Health Platform: Development and Modernization

Healthcare

Success Story Population Health Platform: Development and Modernization image

Efficient Patient Monitoring Ecosystem

Nonprofit & Charity

Healthcare

Success Story Efficient Patient Monitoring Ecosystem image

Thomson Reuters Platform

Pharmaceutical

Healthcare

Success Story Thomson Reuters Platform image

Cloud-based Solutions to Enhance HR and Behavioral Health Platforms

Healthcare

Success Story Cloud-based Solutions to Enhance HR and Behavioral Health Platforms image

AI-powered System: Cybersecurity Report Generation and Risk Mitigation

Healthcare

Success Story AI-powered System: Cybersecurity Report Generation and Risk Mitigation image
01

Contact Us

Accessibility Adjustments
Adjust Background Colors
Adjust Text Colors