Stores faced unauthorized after-hours access with no automated detection. Legacy cameras lacked contextual alerts, manual arming/disarming was unreliable, and constant notifications during open hours overwhelmed staff. Intrusion response times often stretched to 30 minutes, leaving branches vulnerable.
We built a scalable, event-driven system using AWS Lambda, DynamoDB, and IoT Core for secure, serverless cloud orchestration. At the edge, Raspberry Pi devices with door sensors and cameras captured images ±5 seconds around events, processed locally using DeepStack AI for fast and private face detection. Automated scheduling ensured the system was only active during off-hours, eliminating false positives. Slack integration delivered real-time alerts with annotated images, while a React + Spring Boot dashboard enabled managers to control schedules and view logs.
The MVP reduced response times to under 2 minutes, eliminated false positives during business hours, and provided complete visual evidence of all events, preventing multiple intrusion attempts in the first month.
