Portfolio
这里收拢更像作品集的项目：有明确问题、技术路线和个人角色。下面仍保留项目笔记列表，用来记录实现过程中的推导、实验和阶段性想法。
DreamCatch AI focuses on the fragile first minutes after waking: capture speech, structure memory fragments, ask targeted follow-up questions, and generate visual cues that help users revisit dreams later.
Product concept, AI workflow, UX system
A compact vision project for understanding the pre-deep-learning lane detection pipeline: grayscale conversion, Gaussian smoothing, gradient filtering, edge thinning, thresholding, and Hough-space voting.
Algorithm implementation and experiment notes
Build GPT is a from-scratch learning track for demystifying language models: start with tensors and token sequences, then assemble attention, residual blocks, training loops, and sampling.
Model implementation and study route
DreamCatch AI focuses on the fragile first minutes after waking: capture speech, structure memory fragments, ask targeted follow-up questions, and generate visual cues that help users revisit dreams later.
Qwen ASR / LLM structured extraction / FLUXA compact vision project for understanding the pre-deep-learning lane detection pipeline: grayscale conversion, Gaussian smoothing, gradient filtering, edge thinning, thresholding, and Hough-space voting.
Python / OpenCV / CannyBuild GPT is a from-scratch learning track for demystifying language models: start with tensors and token sequences, then assemble attention, residual blocks, training loops, and sampling.
PyTorch / Transformer / Language modelingSearch the grove