January to May 2026 · In production
VITA
Real-time AI voice coaching in English for corporate employees.
My role. Sole engineer: architecture, mobile app, web panel, API and infrastructure.
The coolest part of the project
The student talks to the AI by voice, and the voice never goes through the backend.
The app connects straight to Gemini Live over WebSocket so the conversation has no added latency. The backend only issues ephemeral tokens: the API key never leaves the server. When a session ends, a background job analyzes the transcript and the teacher sees grammar, vocabulary and clarity per lesson.
3 applications
student app, teacher panel and API
5 months
from architecture to production deployment
Prompts in the database
content is tuned without redeploying
How it is built
The pieces and why they are there.
Mobile app
Flutter and Dart, Gemini Live over WebSocket
Web panel
Angular 19 with Signals and TanStack Query
Backend
Django 5, PostgreSQL with pgvector, Celery and Redis
Infrastructure
DigitalOcean, Docker, continuous deployment
Outcome
In production with real corporate students. The client kept hiring new phases after launch.