It watches how people move on security cameras and instantly tells store staff when someone's probably stealing—without needing any new cameras or equipment.

Lexius's core use case applies deep learning-based pose estimation, object tracking, and behavioral sequence modeling to live CCTV feeds to detect shoplifting in progress. The system ingests video streams from existing retail cameras, runs inference on behavioral patterns (e.g., concealment gestures, rapid item-to-bag transfers, prolonged loitering near high-value merchandise, evasive movement toward exits), and generates real-time alerts pushed to staff mobile devices. Critically, the model is trained to recognize movement patterns only—it deliberately excludes facial recognition, skin tone, gender, or clothing as features, ensuring privacy compliance and eliminating racial or demographic profiling. The system continuously learns from confirmed incidents, improving precision over time. Retailers report the system pays for itself within days due to the immediate reduction in theft-related losses, and deployment requires zero new hardware investment.

It's like having a loss prevention officer with perfect attention who never blinks, never profiles, and works every aisle simultaneously—except it's software running on cameras you already own.

Instead of scrubbing through weeks of security footage to find a theft, you just type what you're looking for and the AI finds it instantly.

Lexius's searchable video platform applies multimodal AI—combining computer vision embeddings, temporal event indexing, and likely natural language processing—to convert raw, unstructured CCTV archives into a structured, queryable database. Loss prevention teams can search for specific behavioral events (e.g., "concealment near electronics aisle between 2–4 PM last Tuesday") and receive timestamped, clipped video results in seconds. The system indexes detected events, movement trajectories, and behavioral anomalies across all camera feeds, creating a unified timeline of store activity. This dramatically accelerates post-incident investigations, supports evidence gathering for law enforcement, and enables trend analysis (e.g., identifying repeat offender patterns or high-risk time windows). The approach transforms passive surveillance archives from a storage cost into an active intelligence asset, enabling data-driven loss prevention strategy without requiring manual video review.

It's like Google Search, but for your security cameras—type what happened and it finds the exact moment across thousands of hours of footage.

It learns the invisible patterns of how shoplifters behave differently from regular customers—so stores can put staff in the right place before anything happens.

Beyond real-time detection, Lexius's platform aggregates anonymized behavioral data across stores and time periods to build predictive models of theft risk. By analyzing patterns such as traffic flow anomalies, dwell time distributions, path deviations, and temporal clustering of suspicious behaviors, the system generates heat maps and risk scores for specific zones, aisles, and time windows. Store managers receive actionable recommendations—such as repositioning staff during predicted high-risk periods or adjusting merchandise placement to reduce concealment opportunities. The models are trained exclusively on movement and spatial data, preserving individual privacy while extracting macro-level behavioral intelligence. This shifts loss prevention from reactive (catching theft in progress) to proactive (preventing theft through environmental and staffing optimization). Over time, the system's predictions improve as it ingests more data across the retailer's store network, enabling cross-location pattern transfer and benchmarking.

It's like Waze for shoplifting—it learns where and when the traffic jams of theft happen so you can reroute your staff before the gridlock starts.