ONNX Runtime · WebGPU · React · Computer Vision · Creative Tools
I'm exploring how far modern browsers can go as an AI application runtime. My work focuses on moving useful media and computer-vision workflows from servers to users' devices — reducing uploads, preserving privacy, and making advanced creative tools easier to access.
I enjoy the engineering between a model demo and a real product: ONNX graph optimization, WebGPU/WASM inference, workers, model caching, media pipelines, timelines, and responsive interfaces.
| Project | What it does | Stack |
|---|---|---|
| Timeline Studio | A local-first browser AI video editor with voiceovers, Whisper captions, vision effects, talking avatars, a multi-track timeline, and MP4/WebM export. Live demo → | React, ONNX, WebGPU, WASM |
| Vocal Remover Web | Separates vocals and instrumental audio directly in the browser. Live demo → | JavaScript, Web Audio, ML |
| DeOldify ONNX Web | Browser-based colorization for old photographs. Live demo → | ONNX Runtime Web, HTML |
| Depth Anything Web | Runs quantized monocular depth estimation in the browser. Live demo → | ONNX, Computer Vision |
| EvoArm | An embodied-AI framework for controlling robotic arms with large language models. | Python, Robotics, LLMs |
- Browser-native AI video creation with privacy-friendly local inference
- ONNX model conversion, quantization, and WebGPU compatibility
- Temporal computer vision for video rather than first-frame-only demos
- Reliable browser media pipelines for speech, captions, audio, and export
- Interfaces that turn research models into understandable creative tools
If you're working on browser AI, ONNX, WebGPU, creative tooling, or local-first software, I'd love to exchange ideas. Explore the projects above, open an issue, or find more experiments at ai-creator.top.
Building AI tools that stay close to the user.