About the Company
Our client is a US-based health tech company solving one of the most underserved problems in modern healthcare: the language gap between doctors and patients. Standard translation tools fail in clinical settings - they miss medical nuance, struggle with edge-case languages spoken by immigrant communities (Swahili, Haitian Creole, Azerbaijani, and many more), and introduce delays that are unacceptable in a live consult.
The product is a real-time translation engine fine-tuned for clinical use cases, designed to be safe, accurate, and fast enough to disappear into the background of a doctor-patient conversation.
The Challenge
A real-time, HIPAA-compliant, clinically-fine-tuned voice translation product is one of the harder engineering problems in healthcare AI:
- Real-time voice infrastructure: Sub-second latency from speech to translation, end to end, across noisy clinical environments and edge-case languages.
- HIPAA compliance from day one: Every part of the stack - infra, deployment pipelines, observability, third-party integrations - had to be designed to clear compliance rather than retrofit it.
- Hallucination and accuracy in clinical context: Hallucinated translations can have direct clinical consequences, so detection and intent classification had to be built into the pipeline.
- Multi-channel deployment: In-app, over-the-phone, and dashboard-driven flows had to all work, with telephony partners (3CX, Telnyx, Twilio) plugged in cleanly.
- Operational scale: As the client onboarded clinics and interpreters, the platform had to keep pace with growing transcription minutes per month.
Solution Implemented by Tequity
Tequity has been the client's engineering partner across the entire stack.
Foundational Infrastructure:
- End-to-end HIPAA-compliant frontend, backend, and DevOps pipelines.
- UAT and production deployment environments.
- Observability, alerting, and Vanta items closed for compliance.
Real-Time Translation Pipeline:
- Streaming STT services - migrated and optimized across providers, including Soniox v4, ElevenLabs, and queue-based pipeline overhauls.
- Hands-free in-app and over-the-phone flows powered by LiveKit and 3CX (replacing Telnyx where appropriate), with S3 and webhook setup.
- Multi-language audio diarization to handle conversations that switch languages mid-stream.
- Hallucination detection workflow, intent detection, inactivity-based call ending, and auto-timeout.
- Automated audio-file testing pipeline.
Interpreter Dashboard:
- Built and continuously upgraded the interpreter dashboard application, including UI improvements, analytics, logic updates, and a domain-aware feature flag system.
- Twilio setup and onboarding flows for new clinic clients.
- Department mapping and data cleaning across the platform.
Continuous Improvements:
- Volume boost on outputs, audio worklet optimization, transcription improvements, and a text-translate POC.
- Observability via Slack alerts, UV package management, and LiveKit egress setup.
- Ongoing language additions including Azerbaijani, Swahili, and others, with department-aware prompts.
The Tequity team operates as the client's de facto engineering function across product, infra, and AI.
Results
The platform Tequity built is now live and scaling fast:
- 50,000+ voice minutes of real-time medical translation handled per month
- 30% month-on-month growth in usage
- Multi-language support including edge-case languages (Swahili, Haitian Creole, Azerbaijani, and more)
- HIPAA-compliant infrastructure and operations, with Vanta items closed
