Sentra Retail OS

Sentra Retail OS is a B2B spatial intelligence platform designed to transform standard retail camera infrastructure into a high-precision behavioral analytics engine. As CTO, I architected the system to operate under real-world constraints: limited bandwidth, low-power edge hardware, and the need for 24/7 reliability.

Core Architecture & Technical Highlights:

• Edge-First Processing: Optimized for deployment on low-cost edge nodes, reducing cloud costs by 90%.
• Unified NPU Inference Pipeline: Built a centralized NPU daemon that routes all camera inference through a single Unix socket — eliminating per-process weight thrashing and accelerating frame processing from seconds to ~14ms per frame.
• Optimized Sparse Analytics: A proprietary approach to video processing that maintains tracking accuracy while minimizing CPU load.
• Offline-First Resilience: Integrated Redis-based buffering and load shedding to handle network instability without data loss.
• Multi-Tier Risk Engine: A mathematical scoring cascade (Tier 1-3) for real-time threat detection and behavioral feature extraction.
• Custom Calibration Pipeline: Camera-to-3D store space calibration for accurate shopper trajectory tracking.

Competitions & Recognition:

Sentra has also been validated through a series of startup competitions and pitch events:

• Pizza Pitch — 3rd place
• Aul Hackathon 1 — 1st place
• Aul Hackathon 2 — 2nd place
• Alga Pitch Day — Best Idea
• Infomatrix — 2nd place

Together, these results brought 1.5 million KZT in prize funding for the project.

Website: https://sentra.company