RESEARCH / ENGINEERING / LEARNING
A short path through the work.
从研究代码到工程案例：先看问题与实现，再查看验证方式与局限。
Selected work · 精选项目
- KICP — multimodal research code. CLIP, co-attention, learned prompts and knowledge retrieval for fake-news detection. Start with the repository and external-data requirements; this guide does not add new performance claims.
- Counselor Workflow — an engineering case. Try the role and module views, then inspect the state model, API contracts and verification commands. The public demo uses synthetic interaction data.
- Frontend & Data Lab — learning experiments. Explore the interactive dashboard and multi-step form, then read the Python / SQL pipeline and tests.
Inspect an example · 代码与笔记
- Data labeling quality — complete annotation pairs, agreement, coverage and conflict review.
- Bounded public-web crawler — robots rules, same-origin scope, request budgets and offline tests.
- Model self-study notes — baselines, feature engineering, evaluation, leakage and error analysis.
- Engineering field notes — deployment and capacity-testing decisions, with stated testing limits.
Read the evidence · 阅读方式
- Identify the question, data source and scope.
- Inspect the small implementation or try the demo.
- Follow the repository's reproduction and test commands.
- Read limitations before interpreting a result.
Learning notes describe methods; they are not completed research results. The examples do not establish production throughput or institutional impact.