STARS is a Transformer-based sequential recommendation framework designed for large-scale e-commerce. It addresses cold-start items, diverse user intent, and latency constraints by combining LLM-augmented item semantics, dual-memory user embeddings, context-aware scoring, and an efficient two-stage retrieval pipeline.

LiteVGGT addresses the computational and memory bottlenecks of the Visual Geometry Grounded Transformer (VGGT) for large-scale 3D reconstruction. It achieves significant speedups and memory reductions by introducing a geometry-aware cached token merging strategy that preserves critical geometric information.