Hey! Come and check out my awsome projects here. If you are interested in any of them, feel free to contact me!

AIConfigurator is a unified performance-modeling system that enables rapid, framework-agnostic LLM inference configuration search without GPU profiling. It decomposes inference into analytically modelable primitives and leverages a calibrated kernel-level performance database. Achieves up to 40% improvement for dense models and 50% for MoE architectures, completing searches in ~30 seconds.

This is the internship project I started from scratch solely during Summer 2024. The project was to develop a data platform for FEA simulations. The scope of the project includes building a web application with 6 microservices running on Kubernetes, which connects to OIDC, S3, Postgres, SLURM HPC, email, and some other Apple internal services.
A macOS application that allows users to change the virtual location of their iPhone and iPad without jailbreaking. Built with Electron, React, and go-ios, it supports the latest iOS 26 with USB device tunneling. Features an interactive map interface with AMap and OpenStreetMap integration, allowing users to pick any location worldwide and mock or restore their device's GPS coordinates.

This is the official website of Orka Inc. I used to worked here as a frontend engineer and developed 10+ pages including sign in, payment, FAQ, support, etc.

While working at Orka, I led the architecture design and development of Orkaui library - our own UI developer kit. This project was then maintained by our entire frontend dev team one month after its establishment. Note: the following doc was hosted by Orka Labs, subject to change at anytime.

A visualization project to track user's WeChat and QQ chat history. Millions of chat records are processed to present a realtime and interactive view. Animations are also supported.

This is a web app in coorpporation with BlueCity group Blud for New York University research project "Developing an Automated Online Psychological Intervention for Gay and Bisexual Men in China", where I worked as a full stack software engineer. The service is maintained by NYU IT since 2024.
The study investigates the solutions to the traffic assignment problem in a parallel computing context. We reviewed the existing solutions, shown the performance of our Spark parallel implementation and sketched some conclusions and cast doubts.

We proposed a machine learning pipeline along with several active learning strategies, evaluated several ML algorithms and relevant strategies. Software practitioners will be able to follow this pipeline and evaluate different ML algorithms with our baseline.

(Under construction) During my summer holiday, I developed a website for EventMaker cooperation, a marketing company who makes business in event design and creation and deployed it on a serverless environment.