RESEARCH & VALIDATION

Built on Real
Grocery Experience

Our AI freshness technology was born from hands-on produce floor research and validated through rigorous academic scrutiny at Carnegie Mellon University.

GE

Giant Eagle Field Research

Hands-On Produce Operations Study | Pittsburgh, PA

The Problem Space

Regional grocers face persistent produce shrink challenges. Industry-wide, shrink rates average 6-8%, manual quality checks are subjective and inconsistent across shifts, and markdown timing is often reactive rather than data-driven.

  • • 6-8% industry average shrink rate on produce
  • • 2-4 hours daily on manual quality checks
  • • Inconsistent markdown decisions across shifts
  • • Emerging regulatory pressure (CA AB 660)

Our Research

Our founder spent 4 months working as a Produce Team Leader at Giant Eagle, studying operations firsthand and conducting computer vision experiments to address the challenges he observed every day on the floor.

  • ✓ CV-based shrink detection and early spoilage identification
  • ✓ AI freshness scoring to replace subjective manual checks
  • ✓ Shelf-life estimation for proactive markdown timing
  • ✓ Data-driven placement and rotation strategies

What This Research Became

Every feature in Xpired traces directly back to a problem observed on the produce floor. The studies conducted during this period shaped the product that exists today.

3M+

Training images collected

130+

Produce types supported

<30ms

On-device inference

EfficientNet

B4 architecture

CMU

Technical Validation

Carnegie Mellon University

McGinnis Venture Competition

Expired Solutions was selected as a McGinnis Venture Competition Finalist, validating both the technical approach and business model through rigorous scrutiny by CMU faculty and industry experts.

Training Dataset 3M+ Images
Processing Speed <30ms On-Device
Produce Classes 130+ SKUs
Architecture EfficientNet-B4 + B0

Dual Neural Network Architecture

TwoModelService combining ProduceClassifier (EfficientNet-B4, 135 types) and FreshnessClassifier (EfficientNet-B0) for classification and quality scoring.

CoreML On-Device Inference

Models converted to CoreML for real-time on-device inference on iOS, with server-side PyTorch backend for full-resolution analysis.

GPT-4o Fallback

Budget-controlled GPT-4 Vision fallback for edge cases where the custom model confidence is low, ensuring robust coverage across all produce types.

Estimate Your Savings

See how much your stores could save with AI-powered freshness scoring. Our ROI calculator uses industry data from USDA, ReFED, and EPA.

Try the ROI Calculator

Interested in What We're Building?

We're looking for grocery retail partners to bring AI-powered freshness management to their produce departments.

Get in Touch Learn More