Kunxi Li is a master's student in software engineering at Zhejiang University, expected to graduate in June 2026. He specializes in optimizing large-scale AI models for efficiency and adaptability. His research innovations include a cross-model knowledge migration framework using parameter adapters to bridge heterogeneous architectures, tasks, and modalities; an adaptive KV cache compression strategy for efficient multimodal long-context inference; a hypernetwork-driven LLM compression technique for edge deployment that retains model expressiveness; a backpropagation-free dynamic perturbation method for memory-efficient transformer fine-tuning. Additionally, he developed KV cache optimization for audio-video LLMs through adaptive focusing and cross-modal calibration. A recipient of the National Scholarship and ICPC Asia Regional medals, Kunxi also contributes as a reviewer for top-tier AI conferences.