Challenge
Cloud-based video proctoring platforms suffer from high bandwidth costs, server GPU bottlenecks, and privacy concerns when streaming continuous candidate webcam feeds during large-scale recruitment assessments.
Telix: Edge-WASM AI Proctoring & Recruitment Platform
Combining an Axum + StoneMQ Rust backend with client-side WebAssembly ONNX inference for real-time exam monitoring and video interviews.
- Rust & Axum
- WebAssembly & ONNX
- StoneMQ
- SvelteKit
- Computer Vision
Niten.ID
Approach
We built Telix around a Rust (Axum) microservice and StoneMQ event pipeline paired with a browser-resident Rust WebAssembly video processor running UltraFace (1.1 MB) and YOLO11n (10.7 MB) via tract-onnx. Face tracking, gaze monitoring, and device detection run directly in the candidate browser while structured telemetry events stream to the SvelteKit monitoring console.
Outcome
Delivered 15-30ms real-time face inference in the browser with a 4x memory reduction and 97% lower infrastructure cost compared to server-side cloud vision pipelines.
Results
- 15-30ms client-side face inference in WebAssembly using UltraFace Slim (1.1 MB)
- 97% infrastructure cost reduction by shifting vision inference to the browser edge
- High-throughput event ingestion via Rust (Axum), StoneMQ, and PostgreSQL