Statigest
Retail | Computer Vision
From video analysis to mobile-ready models: computer vision that works on the retail floor.
Statigest provides analytics for retail chains. but their field teams needed a way to analyze shelf placement, product facings, and merchandising compliance in real time. Traditional image analysis was too slow and too tied to server-side infrastructure.
Rakam built a complete computer vision pipeline. from video-based product recognition to ONNX and Core ML model export for iOS and Android. What started as a POC became a production system in 68 days, with models running directly on field teams' mobile devices.
« Statigest is bringing computer vision to the retail floor. from video analysis to mobile-ready models for field teams. »
Statigest
Client partnership
POC
Validated & scaling to production
2.3x
Faster with ONNX
68 d
Integration phase
iOS+
Android deployment
Business
Business Impact
Field teams now analyze shelf placement and product facings directly from their mobile devices, without needing server connectivity. This transforms in-store audits from manual clipboard work to AI-assisted instant analysis.
The progression from POC to production in 68 days demonstrates rapid time-to-value. ONNX optimization delivers 2.3x inference speedup, making real-time analysis practical on consumer-grade mobile hardware.
Product
What We Built
Product Recognition from Video
Custom computer vision model that identifies products directly from video streams. Frame extraction via FFmpeg, product détection, and classification. all optimized for retail shelf environments with varying lighting and angles.
Shelf Overlay & Facing Analysis
Visual overlay system that maps detected products onto shelf planograms, counting facings and identifying compliance gaps. Gives merchandisers instant visibility into shelf execution versus plan.
Annotations Dashboard
Internal tool for annotating training data, reviewing model predictions, and managing the continuous improvement cycle. Enables Statigest's team to refine model accuracy over time with domain-specific feedback.
Core ML / ONNX Export for Mobile
Model export pipeline that converts PyTorch models to ONNX and Core ML formats for on-device inference. React Native integration enables deployment on both iOS and Android, with 2.3x speedup from ONNX optimization.
Technical
Technical Architecture
The pipeline starts with custom PyTorch models trained on retail-specific data, then exported to ONNX for server-side inference and Core ML for iOS deployment. FAISS handles similarity search for product identification.
FFmpeg processes video streams into frames, while React Native provides the cross-platform mobile shell. MLflow tracks training runs and model versions, with all infrastructure running on Kubernetes OVH.
# Stack
Custom PyTorch (training)
ONNX (optimized inference)
Core ML (iOS deployment)
FAISS (similarity search)
FFmpeg (video processing)
React Native (mobile app)
MLflow (experiment tracking)
Kubernetes OVH (orchestration)
# Performance
POC validated, scaling to mobile
2.3x ONNX speedup
iOS + Android ready
Move from AI hype to the real thing
Leave us a message, our team will get back to you within 24 hours.