Building tools to make AI systems see and understand the world better.

Experience in embodied/spatial systems, distributed LLM training, human-LLM interactions, and diffusion image generation.

Has working Chinese fluency, and provides a fast learning curve.

Catch me: 📧 Email | 𝕏 | 小 | 📍 Hong Kong SAR, Shenzhen, and the Greater Bay Area

Selected Works

• How to Convert Audio to Spectrogram Images
  Make audio data visible to vision models.

• Implementing and Optimizing Meanshift Clustering
  Taking advantage of a GPU to speed up meanshift clustering by 6x.

• (Un)successfully Implementing DiffEdit
  A paper from scratch implementation to edit images with text.

• Building SolveIt in SolveIt
  Rebuilding SolveIt, an agentic LLM interface that keeps the human in control, in SolveIt

• Implementing a Neural Network from Scratch
  Progressively reimplementing PyTorch.

• Backpropagation Explained using English Words*
  Explaining backpropagation simply enough so that a middle school student can understand it.

• Implementing Stable Diffusion from its Components
  Implementing Stable Diffusion by piecing together the various models that make image generation possible.

CV

View it here.