SandStar is a global leader in AI retail technology. With a decade of deep dedication to the AI retail track, the company has expanded its business footprint across more than 30 countries and regions worldwide. It long-term serves over 30 Fortune Global 500 enterprises, delivering comprehensive intelligent technology and operational support to nearly 500,000 offline retail stores and terminals.

As the core edge computing foundation for its full-scenario business, SandStar adopts NVIDIA Jetson Orin NX, deeply integrating with a Vision Language Model, NVIDIA NemoClaw reference stack, and advanced AI vision technology. The solution empowers two core scenarios: AI vending machines and AI retail stores. Through years of algorithm iteration and model architecture reconstruction & optimization, SandStar has achieved nearly 40% memory footprint reduction. The system supports dual hardware configurations of 16GB and 8GB: the 16GB version delivers ultimate computing power and ultra-high concurrency performance, while the 8GB version strikes an optimal balance between performance and deployment cost. It flexibly adapts to project positioning from the hardware layer and significantly lowers the overall threshold for large-scale commercial deployment.

AI Vending Machines

Benefiting from the inherent scene semantic understanding capability of vision language models, SandStar automatically analyzes location-based consumption characteristics and user preferences, formulating differentiated product selection and planogram strategies for each site. Combined with historical sales peak curves, the system intelligently calculates and optimizes restocking routes, frequency and time windows. It also supports full-process semantic recognition and real-time early warning against theft, intentional camera occlusion and other abnormal behaviors, building a comprehensive anti-loss defense system and effectively reducing long-term operational costs of retail terminals.

Thanks to memory optimization and upgraded model intelligent scheduling architecture, SandStar supports flexible hardware selection. The 16GB configuration supports ultra-high workload and concurrent inference of multiple models for ultimate fluency and redundant computing power. The 8GB version stably runs core tasks including model inference, real-time AI visual recognition and multi-dimensional business data analysis. Without compromising core service performance, it substantially cuts hardware procurement and O&M costs, providing a dual-path, cost-effective and highly stable deployment option for large-scale global rollout of AI vending machines.

AI for Store Operations

Catering to the full-scenario intelligent upgrading needs of offline chain stores, SandStar takes NVIDIA Jetson Orin NX as the edge computing hub to fully unleash the capability of VLM. Beyond basic functions such as passenger flow statistics, automatic out-of-stock inspection and shelf display compliance verification, it leverages VLM’s outstanding long-time-series behavioral semantic understanding to accurately analyze the full-process execution of staff standardized SOPs, customer in-store movement trajectories, and suspected theft & abnormal behaviors. It breaks the limitation of traditional machine vision, which can only identify isolated single actions, and realizes continuous tracking, semantic decomposition and intelligent judgment of complete behavioral links.

On this basis, the system links core modules including demand-driven ordering & replenishment, intelligent shelf display optimization and data-driven marketing decision analysis. It dynamically generates exclusive operation models for individual stores based on store scale, geographic location, surrounding business landscape and customer portrait features, helping retailers quickly capture local demand changes and flexibly adjust business strategies.

Powered by NVIDIA NemoClaw reference stack, the platform automatically conducts in-depth analysis of multi-dimensional store operational data, and intelligently outputs actionable optimization suggestions covering store operation, scientific manpower scheduling, festival promotion strategies and cross-store inventory allocation. Combined with self-developed AI agents, it builds an exclusive “Retail Operation Expert” for each store. Store managers can conduct real-time interactive consultations to obtain professional guidance tailored to current operating conditions, enabling the standardization and scalable replication of top-tier retail operation experience.

Supported by nearly 40% memory optimization, store intelligent transformation supports flexible hardware configuration. High-end complex retail formats can adopt the 16GB version to support concurrent access of massive cameras and synchronous inference of multiple models for extreme performance and business redundancy. Standard chain stores can deploy the mainstream 8GB version to fully support end-to-end business workloads, including VLMinference, full-scenario AI visual services, comprehensive SOP standard monitoring, and NVIDIA NemoClaw intelligent Q&A & decision generation. In addition, the entire solution reuses 95% of existing in-store cameras, further lowering the entry cost and capital investment for retail intelligent upgrading. With highly universal and replicable technical architecture, it enables rapid mass deployment across global regions and retail formats. Most projects achieve investment payback within 3–6 months, delivering steady performance growth value for retail enterprises in the long run.

Technical Accumulation & Global Deployment Advantages

After a decade of industry accumulation, SandStar has built a core technical team of over 200 professionals, establishing a commodity database covering more than 1.5 million global SKUs and a retail scenario dataset of over tens of millions of hours. Its self-developed algorithm closed-loop reflux system compresses the commercial adaptation cycle of new products from the traditional 7–10 days to only about 10 minutes, highlighting outstanding technological edge in the industry.

The company has built a mature and stable edge-cloud collaborative architecture, deeply integrating NVIDIA NemoClaw intelligent interaction, NVIDIA VSS vision software foundation, together with the powerful edge computing of Jetson Orin NX and self-developed core memory optimization technology. It has forged a stable, elastically scalable AI retail technology base compatible with both 16GB and 8GB hardware specifications.

Adhering to the core philosophy of “Understanding AI, Understanding Retail Better”, SandStar always keeps technology grounded in front-line business workflows, precisely addressing industry pain points including inventory imbalance, data privacy compliance and high intelligent transformation costs. With rich on-site deployment and cross-regional operation experience accumulated in over 30 countries globally, SandStar has formed a set of standardized, reusable and replicable AI retail implementation methodologies, establishing itself as a benchmark service provider driving large-scale, low-cost AI upgrading for the global retail industry.