Computer vision-based, AI-powered solution that enables retailers to make critical operational decisions based on real-time inventory data. It is an essential component for building automated stores, enabling new levels of compliance, shelf availability, and operational excellence.
- Merchandiser takes a picture of a product shelf with his mobile device and the solution gives him instructions for replenishing the shelf: which product should be put in which place, which product is unnecessary on this shelf.
- The same data is used by top management to analyze the efficiency of the entire network through reports by region/city/specific store. Dashboard with analytics for individual categories and recommendations for improving the efficiency of a store or an entire segment.
- The mobile application manages the merchandiser's shift, displays his tasks with detailed instructions, and trains him. Upon completion of the task, the user reports in the same application, having photographed the result of his work, which is also recognized and controlled by AI
WHAT DOES THE SOLUTION LOOK LIKE?
We have developed a SaaS solution based on the use of big data, artificial intelligence and computer vision, consisting of:
- Mobile application for collecting data on shelf status: availability of goods and price tags. The same information can be collected through cameras or robots through predeveloped API.
- The core business logic for processing photos and recognizing products on the shelf based on artificial intelligence, where the magic happens.
- Dashboards for managing store data and employee tasks.
- Dashboards for analyzing aggregated data across the entire retail chain.
Technologies used:
- Python (django in some services)
- Angular
- PostgreSQL, AWS Redshift
- Own AI models, a well-established MLOps process
- PyTorch, TensorFlow, ONNX, TensorFlow Lite, PyTorch Mobile
- Docker, Kubernetes
- AWS, Azure
- Swift
- Kotlin
- Rust
- Grafana
- QuickSight
- Excel ;)
WHAT'S UNDER THE HOOD
Artificial intelligence
- Product recognition by photo with SKU accuracy
- Tool for marking products on photos with shelves for training neural networks
- Text recognition on price tags and on the products themselves (OCR)
- Recognition of people on photos in accordance with GDPR requirements
- Recognition of obstacles blocking shelves with products
Computer vision and image processing
- Recreation the appearance of a shelf using a planogram
- Merging several shelf images
- Straightening images when working with distorting camera lenses
- Determining the distance from a smartphone to a shelf with goods
- Determining the blurriness of a photograph
Dashboards
- System of reports with key indicators for managing retail chain with 10K+ stores
- Administrative panel with data about stores, personnel, products, price tags
Cloud infrastructure
- Cloud infrastructure with cost-saving algorithms on AWS and Azure
- Processing millions of images per day, storing petabytes of data
- QoS 99,5%
Data Administration
- Collecting data on products in a retail chain
- Determining whether a realogram matches a planogram: identifying gaps on the shelf and intervening products
We are engaged
in innovation in creation
and taming of artificial intelligence
in innovation in creation
and taming of artificial intelligence
worked on the project
50+
Project duration
6 yrs
The project was developed by a full-fledged development office: mobile, backend, frontend developers, testers, analysts, designers, ML/CV developers, project/component managers.
