AI Risk Control Platform for Smart Retail

Video data and transaction data connected across about 500 stores, cutting the investigation of a suspicious transaction from a week to a day.

Challenge

A major Japanese cloud camera platform provides smart video management and AI video analytics for retail, manufacturing, logistics, food service, and other industries. To strengthen its digital capabilities for retail, the platform launched a project to integrate its cameras deeply with POS checkout systems. Video monitoring on its own can only "see," and checkout data on its own can only "keep the books." With the two separated, investigating a single suspicious transaction often took a store a week. The project had to connect data from two independent platforms in production, with enterprise-grade stability for about 500 stores running at once.

Solution

Scope of work: The overall backend solution, including APIs, databases, batch processing architecture, multi-system data integration, production monitoring, and DevOps.

Architecture design: Designed the three-layer architecture of APIs, databases, and batch processing, set interface and database standards, and continuously drove API standardization and maintainability improvements.
Multi-system data integration: Designed the data integration between the camera platform and the POS platform. Within about 1 minute of a transaction, it is matched to the corresponding video clip, with video events and transaction records on a single timeline.
Support for anomaly detection: The platform detects anomalous transactions with a combination of rules and machine learning models, covering theft and employee misconduct. The backend delivers the transaction and video data that detection needs, complete and on time, and automates the related business processes.
Development process: Organized requirements analysis, system design, technical reviews, task breakdown, and code reviews, and drove cross-team development to keep the project on schedule.
Production stability and DevOps: Built production monitoring with Datadog and Grafana to support troubleshooting, performance optimization, and stability work, and built cloud-native deployment and infrastructure-as-code delivery with Docker, Terraform, and GitHub Actions.

Results

Live in about 500 stores: The product runs in retail, food service, and apparel stores in Japan and processes about 10,000 transaction records a day, bringing the platform new revenue.
From a week to a day: With transactions and video matched within 1 minute, stores no longer search through footage clip by clip to investigate a suspicious transaction.
7 core APIs, 1 database model, 2 batch systems: Designed and delivered, supporting the integration of video and transaction data.
A reusable method: This architecture and monitoring system for combining "device data + business system data" maps directly to what manufacturers need: "device connectivity → data integration → dashboards and alerts."

More case studies

See other case studies

Eight platform-wide sales events in over two years with zero major incidents, load tested at up to 8 times normal traffic.

A large online produce trading platform, built from scratch and operated, ending in a sale of the business.

On-site deployment for a US robotics AI company, keeping robots running reliably on real construction sites.

The next project to get done should be yours

Write to us about the problem you want to solve. Whether it is worth doing, how to do it, and how soon it can show results: we will give you a professional and direct assessment.