Hi, welcome to my website! I am a Software Engineer at Huawei Technologies, where I work on distributed memory systems and KV cache infrastructure for large language model inference. Previously, I was a Research Intern in the Safe AI Lab at Carnegie Mellon University, instructed by William Han.
I received my Master of Information Systems Management from Carnegie Mellon University in 2025 and my B.E. in Data Science from Shanghai University in 2024.
My research and engineering interests lie at the intersection of AI systems and multimodal AI applications, with a current emphasis on:
(* indicates equal contribution)
arXiv, 2025
TL;DR: We integrate statistical and pathological feature retrieval into ECG-language modeling, substantially improving clinical report generation and diagnostic accuracy.
arXiv, 2025
TL;DR: A unified study of signal, image, and symbolic representations for electrocardiogram-language models.
Springer CUTECSA, 2024
TL;DR: We combine Instant-NGP NeRF with simulated city maps to create an interactive and immersive urban digital twin.
IEEE MLISE, 2022
TL;DR: An applied machine learning pipeline for recognizing and predicting abnormal human behavior patterns.
Huawei Technologies
Distributed memory management and high-performance KV cache storage and transfer for LLM inference; contributor to Ascend MemCache and MemFabric Hybrid.
Carnegie Mellon University · Safe AI Lab
Multimodal AI research on retrieval and reasoning for ECG-language models, instructed by William Han.
Anker Innovations
Real-time heart-rate estimation from multimodal PPG + IMU signals, with model pruning and quantization for efficient inference.
Shanghai Intelligent and Connected Vehicle R&D Center
Real-time vehicle tracking and trajectory analysis across camera networks using DeepStream and YOLOv5.
Meituan & Dianping
Applied XGBoost, unsupervised clustering, and funnel analysis for user segmentation and fraud detection; automated large-scale data preprocessing and anomaly visualization workflows.
Shanghai Jiao Tong University
Applied Instant-NGP NeRF and simulation to build interactive digital twins; combined YOLO and Segment Anything for object detection and segmentation in robotic manipulation.
Industrial Securities Futures
Built Python- and Excel-based portfolio risk analysis tools and automated document parsing and database ingestion, reducing manual processing time by 90%.