Feb 2026 – present

ETITE Meet, a video platform with AI captions and summaries

A multi-tenant video consultation platform on LiveKit, with an embed SDK on npm. It powers telemedicine video calls for Rwanda Virtual Hospital.

For 60 concurrent one-to-one video consultations, with 0.01% packet loss: 1.49 vCPU; For 60 concurrent one-to-one video consultations, with 0.01% packet loss
To transcribe an hour of audio on CPU: ~6.5 min; To transcribe an hour of audio on CPU
Embed SDK with no dependencies: ~2 KB; Embed SDK with no dependencies

The problem

Telemedicine needs video that partner health systems can embed safely, that records reliably and that runs on modest servers. Captions and summaries have to work in Kinyarwanda.

What I built

A multi-tenant API on LiveKit: rooms, recording, live streaming and HMAC-signed webhooks.
A versioned OpenAPI contract with CI drift checks, and an embed SDK published on npm.
An AI service for live captions, consultation summaries and a voice assistant with Kinyarwanda support, now in pilot validation.
Benchmark-driven model choice: the configured speech model would have missed the 20-minute processing target, so I switched to one that transcribes an hour of audio in about 6.5 minutes on CPU.
A load test and capacity model for the telemedicine pilot, with server sizing for 100 to 500 concurrent patients.