01 / Learning platform / Real-time · Salwa Dev
Quran Horizon
A production AI-powered education platform running live one-to-one classes over WebRTC, with a self-hosted speech and language pipeline behind it.
System
case study01
case study01
The problem
Run real teaching sessions reliably, and make the platform understand what happens inside them — without sending a single minute of audio to a paid API.
What changed
- The entire AI pipeline runs self-hosted — speech-to-text, teaching-quality evaluation, homework generation and an in-class assistant, with no paid API spend
- Seven services containerised with Docker Compose behind an SSL nginx reverse proxy, deployed automatically through GitHub Actions
- Browser-based noise cancellation via RNNoise WASM, so audio quality holds up on ordinary home connections
- Live safety monitoring mutes automatically on PII or flagged keywords during a class
Engineering map
Interface — role-based dashboards for students and teachers, multi-language
Real-time — LiveKit video classes, attendance, automated rescheduling
Intelligence — self-hosted Arabic/English/Urdu STT and LLM evaluation
Safety — streaming ASR with PII and keyword detection, automatic mute
Production — seven containerised services, nginx with SSL, CI/CD deploys
Technology
Next.jsLiveKit / WebRTCfaster-whisperOllama (Qwen 2.5)Docker ComposenginxGitHub Actions