Back to Projects

Promixa: AI Transcription Platform

Dual-API Strategy & Cloud-Native Deployment

1. Introduction & Project Goal

Promixa (promixa.me) is an AI-powered platform that delivers high-quality audio & video transcriptions, with future plans for AI image generation. While the UI was built rapidly with the help of AI tools, the project's main focus has always been a robust, scalable backend that showcases modern engineering practices.

2. System Architecture Overview

The system follows a decoupled architecture with separate frontend and backend stacks:

  • Frontend: Next.js (App Router), React 19, TypeScript and Tailwind CSS – deployed on Vercel.
  • Backend: Java 17 Spring Boot 3.4 – handles API, business logic, authentication and integrations with Deepgram & AssemblyAI – hosted on AWS EC2 Free Tier.
  • Database: PostgreSQL on AWS RDS stores users, roles, metadata and usage stats.
  • Security & CDN: Cloudflare manages DNS and protects the public API (https://api.promixa.me) served via JWT cookies.

3. Deployment Journey

The project went through several iterations to achieve a cost-effective but reliable production setup:

  1. Render: Initial proof-of-concept – easy but limiting.
  2. AWS EC2 Free Tier: Migrated for more control; overcame resource limits with optimisation.
  3. Cloudflare: Added for DNS & security hardening.

4. Backend Deep Dive

The backend applies a dual-API transcription strategy – no single vendor excels at every language:

  • Deepgram Nova-3 / Nova-2 for English and general audio.
  • AssemblyAI Direct Streaming for Turkish and automatic language detection.

Spring Security secures the API with JWT (stored in HttpOnly cookies) and Google OAuth2. Role-based guards protect both REST endpoints and React routes.

5. User Management

Features include email/password or Google sign-up, profile editing, a 30-day trial with usage tracking and role-based access control.

6. Conclusion

Promixa demonstrates how a lean team can deliver a production-ready SaaS by combining a powerful Java backend with a serverless React frontend, thoughtful cloud deployment and a pragmatic multi-vendor AI strategy.