Work Experience
-
Machine Learning Engineer, Qualcomm.
Optimized multimodal foundation-model inference and studied accuracy-latency trade-offs
for Stable Diffusion 3/3.5, Qwen-VL/Omni, and Phi-4-Multimodal. Developed full-stack
adaptations across model structure, quantization, operators, speculative decoding, and
deployment pipelines, validated through on-device benchmarks.
-
High Performance Computing Engineer - AI Inference, DJI.
Optimized heterogeneous inference for autonomous-driving BEV and occupancy networks through
preprocessing, custom operators, kernels, compiler passes, and hardware-aware pipeline design
across Qualcomm CPU, GPU, and DSP backends.
Research Experience
-
Empirical Study of Conditioning Structure for Efficient MoT-based Multimodal Models.
Mar 2026 - Present. Advisors: Dr. Yuexiao Ma and Dr. Xiawu Zheng. Profiles visual and text
conditioning signals in BAGEL-7B-MoT across CFG branches, roles, layer depths, and denoising
timesteps.
-
AlphaQ: Calibration-Free Bit Allocation for MoE Quantization.
Aug 2025 - Present. Advisors: Dr. Shiwei Liu and Dr. Yuexiao Ma. Develops calibration-free
expert and layer importance estimation from spectral structure and solves global mixed-precision
bit allocation with integer linear programming.
-
Research Assistant, TU Darmstadt AI & Machine Learning Lab.
Oct 2021 - Feb 2022. Designed a cloud-based experimental environment for studying
learning-agent behavior through multi-agent interactions and ran more than 300 evaluation
experiments.
Education
- M.Sc. Computational Engineering. Technical University of Darmstadt, Germany, Sep 2020 - Sep 2023.
- B.Eng. Mechanical Engineering, Robotics Specialization. Beijing Jiaotong University, China, Sep 2015 - Jul 2019.