I build practical AI tools and contribute focused fixes to open source.
My interests span computer vision, local inference, and developer tooling.
Software Engineering & Artificial Intelligence · Zhejiang University of Science and Technology · Hangzhou
Explore my projects · View my pull requests
Selected work accepted upstream, spanning performance, annotation workflows, and regression testing.
- Image loading & AI caching — Avoid full-image copying and hashing during loading, while preserving embedding reuse when revisiting unchanged images.
- Large-image handling — Reject images that exceed Qt's decoding limits before decoding pixels, avoiding unnecessary work and memory use.
- Annotation workflow — Double-click a shape to edit its label, with coverage for selection and overlapping shapes.
- Interaction testing — Exercise linestrip hover and point insertion through real Qt events, guarding against insertion on an invisible closing edge.
- Development dependencies — Remove an unused Twine dependency and its transitive packages.
| Project | Contribution |
|---|---|
| Kornia | Support integer-valued custom colormaps, preserving the default floating dtype and explicit dtype overrides. |
| CatGPT-Gateway | Add concurrent browser tabs, persistent sessions, and SSE streaming. |
| Project | What it does |
|---|---|
| WebLLM Pro | A browser interface for local LLM inference with WebGPU, model caching, and persistent conversations. |
| Codeforces Analytics | A browser extension for visualizing competitive-programming activity and exporting analytics. Archived. |
Python · PyTorch · NumPy · JavaScript · WebGPU · Qt · Git