AptStock: AMD-Accelerated Inventory Intelligence

AptStock — AI Demand Forecasting & Inventory Optimization on AMD AptStock transforms point-of-sale (POS) history into demand forecasts and inventory decisions for Multi-SKU stores—even when current-stock data is unavailable. Multi-SKU stores can upload simple CSV/Excel sales data to explore product-level forecasts, forecast ranges, stockout risks, safety-stock estimates, reorder triggers, and recommended stock for 7-day and 15-day planning. AptStock prioritizes replenishment, identifies fast-moving products and slow/dead stock, estimates remaining stock coverage when inventory data is available, and generates purchase lists and exportable reports. Its voice assistant helps retailers identify products needing attention and decide what to reorder first. For AMD Developer Hackathon: ACT III, AptStock uses PyTorch and AMD ROCm to run batched inventory calculations on AMD Instinct MI300X hardware. The workload includes demand statistics, safety stock, 7-day and 15-day stock recommendations, and reorder points. Input validation, CPU-reference comparisons, and shadow-mode execution help verify numerical consistency while preserving the CPU-authoritative recommendation path. In the recorded sample-route integration test, CPU and AMD inventory outputs matched for 5,000 SKUs. An end-to-end benchmark tested 500, 2,000, and 5,000 SKUs using 180 days of synthetic demand history. Timing included data transfer, GPU computation, and result transfer. All nine inventory outputs passed numerical-tolerance checks. The October 18, 2026 report recorded a 1.446× CPU-to-AMD speedup at 5,000 SKUs. This result applies to the tested configuration, not every workload or the complete forecasting pipeline. AptStock combines practical retail decisions, explainable inventory mathematics, conversational insights, AMD execution, CPU/GPU validation, and transparent benchmarking. Next steps include broader retailer-data validation and measurement of forecast quality and operational outcomes.