Frequently Asked Questions

Our food detection model is trained on institutional dining imagery and validated across cafeteria, tray-return, and composting workflows. Real-world accuracy depends on lighting, menu, and tray-return layout, so we measure it per site during onboarding rather than publishing a single benchmark score. Accuracy continues improving with each new pilot environment.

The system is trained on institutional dining food types including entrees, sides, produce, dairy, and beverages. We continuously expand coverage based on pilot menu data.

Yes. CPIsight is designed to mount over plate return stations, tray conveyors, or composting drop points. We configure the viewing angle and detection zone during onboarding.

Our model estimates portion remaining using depth sensing (Intel RealSense) + visual inference. It handles partial plates, mixed items, and scraped trays — and correlates disposal data back to the original serving event.

Typical installation is 2–4 hours. Our hardware mounts above the plate return station with no structural modifications required. Most sites are operational the same day.

Minimal IT involvement. CPIsight runs edge inference on the device — no cloud dependency for detection. It does require a standard WiFi or ethernet connection for data sync. We handle all configuration.

We offer data export via CSV and API. Direct POS/ERP integration is on our roadmap. Contact us to discuss your specific system.

Each CPIsight unit includes the Jetson Orin Nano Super edge compute module, Intel RealSense D415 depth camera, mounting hardware, and power supply. Everything ships pre-configured.

The camera is mounted overhead at the tray-return station and aimed down at trays — it is positioned to capture food and trays, not people. There is no facial recognition of any kind, and waste data is never linked to any individual student or diner. Because it watches a live tray line, incidental imagery (such as a hand placing a tray) can appear at the edge of frame; we are formalizing image retention limits and a privacy verification program as part of our K-12 compliance work.

Waste event data is processed on-device and synced to your institution's secure tenant in our cloud (AWS S3 + Supabase). Data is fully isolated per institution using row-level security. You own your data.

We are actively working toward FERPA compliance for K-12 deployments (formal privacy verification is in progress) and can discuss HIPAA considerations for healthcare partners. The system is designed to avoid student and patient PII: no facial recognition, no linkage between waste events and individual identities, and aggregate-only analytics. Contact us for a compliance discussion specific to your institution.

Yes. Full CSV export is available from the dashboard, and we offer API access for deeper integration.

Yes. CPI works directly with school districts and institutions today, whether dining is self-operated or FSMC-managed — the institution is our customer and owns its data. Longer term we see FSMC partnerships as the preferred path, and we're glad to include your management company in the pilot conversation.

We handle everything: hardware shipping, installation coordination, model tuning for your menu, staff orientation, and dashboard setup. Most pilots are live within 2 weeks of contract signing.

You get a dedicated point of contact, access to our support portal, and proactive model updates. We monitor system health remotely and alert you to any issues before they impact your data.

Yes — pilots are our preferred starting point. We believe the data should speak for itself. Book a demo to discuss pilot terms for your institution.