Hey y’all!✨
I’m Jiayin, a rising second-year Master’s student in Computer Science at the University of California, San Diego, advised by Prof. Tzu-Mao Li. I received my B.S. degree in Computer Science, with a minor in Physics, from the University of Illinois Urbana-Champaign. I’m broadly interested in computer graphics as a way to reconstruct, simulate, and reinterpret the physical world through computation. My current focus lies in rendering, reconstruction, and scalable graphics systems. I’m also interested in computer vision, XR, and robotics, especially where they intersect with graphics.
I’m always open to connecting and sharing ideas. Please feel free to reach out!
Education
Sep 2024 – Jun 2026
M.S. in Computer Science, GPA: 3.97
Jan 2021 – May 2024
B.S. in Computer Science, minor in Physics, GPA: 3.95
Experience
Jun 2024 – Sep 2024 | Seattle, WA
Software Development Engineer Intern
- Designed and built an automated benchmarking framework, reducing a 1-2 day manual setup process to a 2-3 minute one-command execution, saving significant engineering time.
- Developed a resource cleanup handler to simplify teardown of benchmarking environments, streamlining post-test workflows.
- Technologies: Python, Lambda, Step Functions, CDK, EC2
Jun 2023 – Aug 2023 | Seattle, WA
Software Development Engineer Intern
- Developed a diagnostic library for Network Load Balancers that resolved noisy mismatch issues in dependency auditing systems, cutting false positive rate by 80%.
- Implemented intelligent mismatch snapshot adjustment via Amazon S3 client APIs, enabling customizable time windows for more accurate and timely debugging.
- Technologies: Python, Amazon S3, EC2, SQL
Projects
Keep your ARMS and LEGS: Assuaging VRAM and Training Speed constraints in Language Embedded 3D Gaussian Splats [3DGS, CLIP, DINO]
Modified the FMGS pipeline to reduce VRAM usage and training time in open-vocabulary 3D scene understanding. Our approach replaces MLP-based decoding of multi-resolution hash encodings with CNNs applied to rendered feature fields, achieving 37% faster training and 24% lower memory usage while maintaining accuracy on the LERF benchmark.
Volumetric Path Tracing [C++, Monte Carlo, Participating Media, MIS]
Implemented a volumetric path tracer supporting absorption, scattering, heterogeneous volumes, and next event estimation. Features null-scattering and spectral extinction handling, integrated into the Lajolla physically-based renderer.
Disney BSDF [C++, Monte Carlo, Path Tracing, BSDF]
Implemented the Disney principled BSDF in a Monte Carlo path tracer (Lajolla). Combined microfacet-based BRDFs with importance sampling to support a wide range of realistic materials, including metal, glass, clearcoat, and retroreflective fabrics.
Ego-Agent Trajectory Prediction in Argoverse 2 [Python, PyTorch]
Built an ADAPT-inspired deep learning model for ego-agent trajectory prediction in a Kaggle competition using a modified Argoverse 2 dataset. The model captures temporal and social context through LSTM and attention modules, and generates long-horizon forecasts via coarse-to-fine endpoint conditioning.
Crystal Quest [Unreal Engine, Blueprint Visual Scripting]
Crystal Quest is a 3D platformer designed with a progression-driven level structure and intuitive visual guidance. Built in Unreal Engine using Blueprint, the game challenges players to navigate dynamic environments through stealth, timing, and light combat mechanics.
Misc
I enjoy photography, hiking, stargazing, playing pool and chess, and hanging out with my cat, Yuki😼. I also like watching movies and reading in my free time.
📸 My photography: @mizoreto
🧶 My cat: @_yukiiii_i