Vosíme Pizza Checker
End-to-end computer vision system for automated pizza quality control with custom hardware and a full React admin.
Custom smart camera hardware (Raspberry Pi, PiPoeCam OS) installed above the pizza oven captures every pizza. A Python cloud backend runs ML quality evaluation in real time. A React admin gives the kitchen team a live dashboard with order tracking, quality stats, AG Grid tables, image zoom, and Leaflet store maps.
Client: Vosíme

- Hardware
- Custom PiPoeCam units
- Evaluation
- Real-time ML quality check
- Coverage
- Multi-location dashboard
- Review
- Traceable evidence
The challenge
Vosíme operates multiple delivery kitchen locations producing high pizza volume. Manual quality checks were inconsistent - pass/fail decisions depended on individual staff judgment and varied by shift. Failed pizzas that slipped through led to customer complaints and costly re-orders. There was no systematic record of quality events per order or per location.
Our solution
We designed and deployed a complete end-to-end system. Hardware: custom PiPoeCam units (Raspberry Pi + PoE camera) running PiPoeCam OS - a minimal custom Linux image with WiFi config and tools for stable reusable deployment - installed above each oven. Software: a Python backend receives images, runs ML quality evaluation, stores results, and publishes events. A React/TypeScript admin gives the operations team branch- and date-aware dashboards, order and pizza tables, capture-quality diagnostics, annotated image review, component-level evaluation, and camera administration. CI/CD uses GitHub Actions with Kubernetes deployment via Helm charts.
Result
The current production interface connects operational monitoring with traceable per-pizza evidence. Teams can review capture quality, detection, ingredients, extras, removals, doneness, and photo-quality checks from a unified multi-location application.
Delivery scope
What we built
- Custom PiPoeCam hardware: Raspberry Pi + PoE camera with custom OS image
- PiPoeCam OS: minimal Linux build with WiFi config, stable reuse across units
- Python ML backend: real-time quality evaluation per pizza image
- Vosíme Admin: React/TypeScript dashboard with AG Grid order tables
- Image zoom with quality annotation overlay
- Leaflet multi-location store map with per-location quality stats
- Branch- and date-aware operational views with location comparison
- Capture-quality diagnostics for empty, obstructed, and evaluable images
- Component-level review of ingredients, extras, removals, and doneness
- Semantic versioning with changelog (v0.0.98 observed in production)
- GitHub Actions CI/CD pipeline, Kubernetes deployment via Helm charts
- Prometheus metrics + Grafana observability
Approved interface evidence
Visuals



Your operating problem



