🎓 M.S. in Statistics (Financial Engineering) @ USTC · Class of 2027

like-attract

|

I build LLM-powered tools, agentic workflows and data pipelines — and ship them end to end.

🚀 Featured: video-to-note GitHub Profile

Featured Project

Independently designed, built, tested and shipped — from product idea to portable release.

VideoToNo ● v1.3.5 shipped

Video → timeline-annotated Markdown notes. A local-first service: feed it a Bilibili / Douyin / YouTube link or a local file, it pulls platform subtitles (incl. Bilibili AI subtitles) or transcribes offline with faster-whisper, then an LLM writes structured, time-stamped notes.

  • Built-in MCP service (stdio / SSE) exposing 10 tools + Agent Skill — callable by Codex, Claude Code & other AI clients
  • Multi-LLM: DeepSeek / Qwen / GLM / Kimi + any OpenAI-compatible endpoint, with multi-gateway fallback
  • Long-video pipeline: chunk → compress → hierarchical merge; timestamp-coverage guardrails, resumable tasks
  • Engineering: FastAPI async backend, pytest + GitHub Actions CI, PyInstaller portable build
Repository ⬇ Download (Windows)
VideoToNo main UI Generated notes

More Projects

Smaller tools and notes — open by default.

🧩

deepseek-harness-desktop

A modern desktop client for the DeepSeek Harness (DSH) plugin ecosystem — everything is a plugin, including the desktop itself.

Desktop · Plugin system
🔭

pi-multivision

pi extension: native vision tool with multi-backend fallback (Step-3.7-Flash → GLM → Qwen) for image understanding.

LLM tooling · Vision · Fallback chain
📝

leetcode-hot100-notes

Notes on the Hot 100 problems I got wrong, over-thought, or missed the key insight on.

Algorithms · Interview prep

Experience & Education

From quantitative finance to LLM-driven data engineering.

Suzhou Institute of Intelligent Computing Technology, CAS (IICT)

Financial Data Analysis Intern · 2025.06 – 2025.11

Built an automated pipeline over 4M+ multi-source records across 8 categories: LLM semantic classification → Elasticsearch vector retrieval → industry-chain knowledge-graph mapping → MySQL. Multi-model cross-validation (ChatGPT / GLM / DeepSeek) with prompt engineering; Python multithreading + checkpoint-based retries for month-long batch jobs; local deployment of BGE-M3 / GPT-OSS to cut commercial API costs.

Shanghai Weijing Information Technology

Quant Research Intern · 2024.05 – 2024.08

Developed & tested government-bond / policy-bank-bond yield-curve models (2022–2024) and Z-Spread pricing; MA / ARIMA / Holt-Winters time-series models with adaptive tuning, extended across multiple bond types.

University of Science and Technology of China (USTC)

M.S. in Statistics (Financial Engineering) · 2024 – 2027

GPA 3.98 / 4.3 · TA for Data Mining (Python) · Academic Scholarship (First & Second Class)

Shanghai University of Finance and Economics (SUFE)

B.S. in Economic Statistics (Accounting minor) · 2019 – 2023

Econometrics 97 · Machine Learning 94 · Mathematical Statistics 90

🥇 MCM/ICM — Meritorious Winner (2021)

Improved PageRank over a 5,600+ node musician-influence network; PCA + cosine-similarity genre similarity matrix.

🏅 CUMCM — Shanghai Division, Second Prize (2020)

Logistic credit-risk default model + nonlinear programming for SME lending terms.

Tech Stack

Tools I reach for daily.

PythonFastAPIVanilla JSElasticsearchMySQLLinuxMCPRAGPrompt Engineeringfaster-whisperpytestGitHub ActionsPyInstallerPandas / NumPy