Hi HN, we’re the creators of Borr AI. We’ve open-sourced a pipeline designed to turn physical retail environments into structured, auditable data. In digital products, we have high-fidelity analytics for every user interaction. In physical retail, that data is usually locked inside proprietary black boxes or grainy CCTV. Borr is a self-hosted platform that fuses stereo vision and weight telemetry to create a "computational truth" for physical spaces. The Use Cases: Forensic Theft Detection Instead of just recording video, Borr generates a hashed "evidence packet." It correlates 3D proximity events with weight sensor deltas to provide a mathematical certainty estimate of an interaction, making it auditable for security and court admissibility. We provide the primitives to track customer journeys in 3D without cloud facial recognition. You can analyze shelf "dwell times," product interaction rates, and heatmaps with millimeter precision using our DLT (Direct Linear Transform) solvers. Checkout-Free Retail: The system could be expanded to track "identity custody" across camera blind spots using a BIP solver. This allows developers to build "Just Walk Out" experiences where transactions are validated by the fusion of vision and weight data. Technical Stack: Dual-camera triangulation using YOLOv8-pose. Probabilistic correlation of weight sensors (0.7 weight) and vision events (0.3 weight). Locally-Aware Transformers (LATransformer) to maintain persistent user IDs without biometrics. |