Chun-Yi Lee

Chun-Yi Lee(李俊毅)

Data Scientist · Computational Neuroscientist

National Taiwan University & Academia Sinica · Taipei, Taiwan

我把處理高維度、高雜訊、多模態資料的方法,從神經科學帶到製造、金融與語言領域—— 引用溯源的 RAG、多代理人 LLM 系統、中文對話分析、電腦視覺特徵萃取、半導體良率預測;也從後端到部署維運, 交付過正式上線的國際會議報名與金流系統。十年 fMRI 與行為資料的訓練, 讓我習慣在高雜訊資料上做嚴謹的評估。2024–2025 年以教育部公費在美國波士頓擔任 Brandeis University 訪問學者近一年,日常以英文與國際團隊協作,跨文化溝通自在。

目前為台大與中研院跨領域神經科學博士候選人,預計 2027 年 3 月畢業, 正在尋找台灣的資料科學、機器學習與 AI 工程職缺。

工作經歷

教育背景

  • PhD
    Interdisciplinary Neuroscience
    National Taiwan University & Academia Sinica
  • MA
    Graduate Institute of Brain and Mind Sciences
    National Taiwan University
  • BS
    Forestry and Resource Conservation
    minor in Life Sciences
    National Taiwan University

核心能力

程式語言

  • Python
  • R
  • MATLAB
  • SQL
  • Shell
  • TypeScript
  • JavaScript

機器學習

  • Supervised Learning
  • Classification & Regression
  • XGBoost
  • Graph Neural Networks
  • Time-Series Analysis
  • Anomaly Detection
  • Imbalanced Classification
  • Concept Drift
  • Information-Theoretic Features (Entropy, JS Divergence)

AI 與 LLM

  • RAG (Retrieval-Augmented Generation)
  • Embeddings & Semantic Search
  • Hybrid Search (Dense + BM25, RRF)
  • Prompt & Context Engineering
  • Multi-Agent Systems
  • Local LLM Deployment (Ollama)
  • Knowledge Graphs
  • NLP / Text Classification
  • Chinese NLP
  • Claude Code

LLM 評估

  • RAG Retrieval Metrics (Hit@k, MRR)
  • Citation Grounding & Precision
  • Hallucination Evaluation
  • LLM-as-Judge & Judge Validation
  • LLM Data Labeling with Human Validation

電腦視覺與訊號處理

  • OpenCV
  • YOLOv8
  • HOG Features
  • Object Detection
  • Signal Processing
  • Multimodal Feature Extraction

資料工程

  • Feature Engineering
  • Data Preprocessing & QC Pipelines
  • Pipeline Development
  • Cross-Site Data Integration
  • Database Management (Shell/SQL)
  • Vector Databases (ChromaDB)
  • APIs & Webhooks (FastAPI)
  • Scheduled Pipelines & Automated Reporting

Web 開發

  • Next.js / React
  • NestJS (Node.js)
  • PostgreSQL & TypeORM
  • OAuth Login (Google)
  • Payment Gateway Integration
  • Docker & Nginx Deployment
  • CI/CD (GitHub Actions)

評估與統計

  • Experimental Design
  • Statistical Modeling
  • Model Calibration (Brier Score)
  • Leakage-Safe Temporal Splits
  • Confidence Intervals (Wilson, Block Bootstrap)
  • LLM–Human Label Agreement (Cohen's Kappa)
  • Synthetic Benchmarks with Planted Ground Truth
  • Ablation Studies

量化研究

  • Walk-Forward Backtesting
  • Information Coefficient (IC)
  • Survivorship-Bias-Free Universes
  • Sector-Neutral Portfolios
  • Overfitting Controls (Out-of-Sample, Deflated Sharpe)
  • Pre-Registered Forward Validation

研究方法

  • fMRI (Functional Magnetic Resonance Imaging) Data Analysis
  • Virtual Reality (VR) Experimental Design
  • Cross-Cultural / Cross-National Study Design
  • Behavioral Data Processing