Hi, I’m Zijie.
I’m a programmer, problem solver, researcher, and data scientist.

About me

A little about me

Learning by building.

I’m majoring in Computer Science and Data Science at the University of Michigan, with interests in AI, machine learning, and data engineering. I enjoy experimenting with models, building data pipelines, and turning ideas into useful tools.

Outside of research and programming, I enjoy photography and badminton, learning guitar, and probably a little too much milk tea.

My photography

Experience & research

01 / 05 Experience

GEMS Lab

Research Intern

March 2026 — Present

Ann Arbor, MI

Reimplemented MD-Syn in PyTorch without published code to reproduce baseline results for drug–drug synergy prediction, achieving 0.91 AUROC on DrugCombDB.
Conducted ablation studies across 1D and 2D feature extraction modules, finding that 1D features accounted for approximately 99% of AUROC performance, while adding 2D features improved recall by 2.4 points.
Implemented Tucker decomposition to generate drug–protein graph embeddings, improving AUROC by 0.02–0.07 over the identity-only baseline on in- and out-of-distribution drug splits.

DrugComb drug–protein graph and connection statistics

02 / 05 Experience

Michigan Institute for Data and AI in Society

Software Developer

Aug. 2026 — Present

Ann Arbor, MI

Updated a federal grant tracker used by 675+ affiliate faculty members, adding 1,100+ opportunities and improving pipelines that process data from Grants.gov, SAM.gov, and the Federal Register.

MIDAS AI and Data Science Funding Digest grant tracker

03 / 05 Experience

Innovation for Impact

Software Developer

Jan. 2026 — Present

Ann Arbor, MI

Built a chemotherapy dose scheduling web app for Michigan Medicine using Flask and a SQLite + JSON database, collaborating with a cross-functional team to replace pen-and-paper workflows and improve efficiency.

Michigan Medicine chemotherapy dose scheduling application

04 / 05 Experience

ARTIST Group

Research Assistant

May 2025 — Dec. 2025

Ann Arbor, MI

Reduced manual data processing time by 70% with a Python pipeline that automated cement plant-to-city transport distance calculations using the OpenRouteService API.
Optimized routes to cut transportation carbon emissions by 16%, implementing a minimum-cost flow algorithm to model cement distribution.

Cement distribution dashboard showing shipment routes and flow volumes

05 / 05 Experience

Computer Vision Lab

Research Intern

May 2025 — Aug. 2025

East Lansing, MI

Built a PyTorch evaluation framework using MaskDINO to compare object-masked and full-image depth estimation metrics, revealing model bias toward foreground objects over global scene geometry.
Evaluated Metric3D and UniDepth across distance ranges on KITTI and DDAD benchmarks using NumPy and OpenCV, finding a 15–20% accuracy drop beyond 30 meters.

Metric3D examples showing source images, depth estimates, and surface normals

Selected projects

01 / 04 Generative AI

Instruction-Tuned Diffusion Model for Image Editing

Fine-tuned Stable Diffusion with LoRA for text-conditioned image edits, using classifier-free guidance on both text and source images to control edit strength and prompt adherence.

Trained on the Great Lakes HPC cluster for 50+ hours on 250k+ instruction–image pairs.

Python · PyTorch · Stable Diffusion

View project on GitHub