Shuai Zhang

I lead the development of on-device agentic AI pipelines at Qualcomm — perception, reasoning, planning, and tool use — deployed across smart glasses, AI PCs, and agentic IoT systems. My team builds the full stack: multimodal perception (segmentation, detection, depth), VLM/LLM reasoning, retrieval and memory, and efficient on-device inference under tight latency and power budgets. Over 8+ years at Qualcomm, I've shipped flagship computer-vision features that reach millions of devices. Limitless Segmentation on Snapdragon 8 Elite, our Segmentation-based Cognitive ISP (winner of the 2023 Edge AI & Vision Product of the Year Award), and our personalized multimodal AI for smart glasses are featured across Qualcomm's developer ecosystem and major industry events including MWC, CES, and CVPR. I'm the leading and primary organizer of the IEEE Low Power Computer Vision Challenge (LPCVC) and the CVPR Workshop on Efficient Deep Learning for Computer Vision (ECV) for both 2025 and 2026, where I'm helping Qualcomm build an edge-AI developer ecosystem now exceeding 3,000+ members. LPCVC 2025 brought together 77 teams from 25 countries — featured in the IEEE Computer Society's interview report. My research spans real-time segmentation, model optimization, efficient transformer architectures, and most recently agentic multimodal systems. I've authored 24+ papers in top venues including ICML, AAAI, ICLR, KDD, IJCAI, and SIAM journals — with my KDD 2020 paper AutoShuffleNet selected as a Most Influential KDD Paper — and hold 18 granted patents in computer vision and on-device AI. I earned my Ph.D. in Mathematics from UC Irvine under Prof. Jack Xin, with a foundation in numerical optimization and compressed sensing. Areas I'm most excited about right now: · On-device agentic AI and VLA (Vision-Language-Action) systems · Multimodal perception for embodied AI — rob