Under the hood: how Voxa AI is built

Voxa is split into a Next.js frontend and a FastAPI backend. The frontend is deployed on Vercel; the backend runs on Render and stores data in a serverless PostgreSQL database on Neon.

The backend follows an MVC-style layout: controllers define the API routes, models describe the database tables with SQLAlchemy, and services wrap external providers.

All AI features — chat, tutor, notes, presentations and voice — go through one AI service. It speaks the OpenAI-compatible chat API, so it can use OpenAI or Google Gemini by changing configuration, with automatic fallback when a model is busy.

Authentication uses bcrypt-hashed passwords and short-lived JWT access tokens. Rate limiting with SlowAPI protects the public endpoints.

Keeping providers behind a single service layer made it easy to switch models without touching any feature code.