ML Vision

Our platform uses real-time facial recognition to automate attendance, improve accuracy, manage students across classes, and generate instant reports with confidence scores

Comment

Summary
Creative ML-vision stack that ingests raw hyperspectral frames, applies GPU-accelerated face-recognition pipelines, and orchestrates real-time inferences across cloud-native microservices with automated deployment topology—accuracy reimagined.

Inspiration
We set out to reclaim classroom time by replacing manual roll call with a deterministic, AI-driven attendance pipeline hardened for enterprise schedules.

What it does
ML Vision streams live video, performs low-latency facial embeddings, reconciles detections against roster services, and emits signed attendance events plus analytics-ready reports.

How we built it
A Spring Boot microservice mesh, React SPA, and Python OpenCV/face-recognition pipeline communicate over REST and WebSocket channels, sharing session metadata to process camera frames in near real time.

Challenges
Synchronizing Python-side recognition events with Java persistence, tuning end-to-end latency on high-frame-rate feeds, and stabilizing cosine thresholds under adversarial lighting conditions required multiple profiling passes.

Accomplishments
Shipped an end-to-end system featuring continuous detection, roster CRUD workflows, encrypted object storage, and exportable attendance reports with audit metadata.

What we learned
Mastered cross-language contract design, face-recognition optimization (batching, quantized models), and jitter-free streaming in constrained classroom networks.

What’s next
Next iterations target adaptive accuracy tuning, native mobile capture surfaces, multi-camera federation, and richer reporting/alerting pipelines.

Share this project:

Updates