AI product prototypes · SQL analytics · Reproducible research
I turn business questions into product workflows, working prototypes, and analyses that can be inspected and tested. My projects explore where AI helps, where human judgment belongs, and how to verify the result.
我关注 AI 产品与数据驱动决策：把业务问题拆成可运行的流程，用清晰的评估和可追溯的证据检验结果。
Explore my AI product portfolio · 查看产品案例集 →
| Project | What it demonstrates |
|---|---|
| AI Product Portfolio | Three product case studies: video preproduction, evidence-based intelligence, and guided mathematics learning. Includes screenshots, a demo video, and a Chinese PDF portfolio. |
| AI Video Preproduction Agent | A workflow from brief and creative concepts to scripts, storyboards, review, and a versioned delivery package. Human decisions and traceable revisions are explicit. |
| SafeDrive SQL Analytics | An 11-table MySQL model and 10 business queries for a fictional designated-driver service: revenue, repeat purchases, service quality, and unpaid orders. |
| Credit Default Prediction | A four-model comparison connecting validation, decision thresholds, and an explicit business cost matrix. |
| Signal Research Framework | A reproducible pipeline for candidate signals, look-ahead checks, backtesting, risk gates, and walk-forward evaluation on synthetic data. |
| Campus Food Pickup Simulation | A discrete-event simulation comparing pickup policies with repeated runs and uncertainty estimates. |
- SQL 交互研究台账：从 10 个业务问题追到订单证据，比较边界条件、保存笔记和导出结果。公开演示使用合成数据；页面结果为 Python 推导，已通过 MySQL 9.7.1 两库实跑对照、触发器与事务回滚验证。
- Choice 置信度工作台：查看不同置信度阈值下的放行、人工复核和已标注错误，保存可复查的分析报告。
- 表格问数与证据工具：围绕 Excel 原生表格回答问题，保留计算口径、单元格证据和源文件变更校验，附可运行的合成订单示例。
- Define the decision: identify the user, the business question, and the outcome to evaluate.
- Design the workflow: make inputs, state changes, human decisions, and failure cases explicit.
- Build and verify: connect prototypes or analyses to inspectable outputs, tests, and documented limitations.
Tools used across these projects include Python, SQL/MySQL, PostgreSQL, FastAPI, Next.js, and statistical and machine-learning methods.
China equity fund evaluation · Courier action classification · Index futures pairs research · Oregon health insurance experiment replication
The AI product cases describe prototypes and their documented validation, not claims of production adoption. Several research repositories are public reconstructions of coursework or research workflows; synthetic demonstrations are labelled, and restricted source data are excluded. Each repository explains its scope, data, and limitations.
AI-assisted implementation is part of my workflow. My focus is on problem definition, product and analytical choices, review, and verification; project-specific attribution is documented in the repositories.