Resume

Experience across applied ML, full-stack AI, and AI research.

A web version of my resume, with a downloadable PDF available for sharing.

Profile / Curriculum vitae

Sky Lee

Applied ML · Full-stack AI · AI Products

[email protected] San Diego · Taipei · Tokyo

Classmethod

Jul. 2026 - Sep. 2026

Tokyo, Japan

Developed Physical AI applications for Unitree G1 and Go2 robots, spanning real-time visual inspection and robotic manipulation.
Built and optimized a visual inspection pipeline using Unitree SDK2, CycloneDDS, and Grounding DINO, reducing overlay latency by 92% and improving image updates from 3-4 seconds to approximately 1 second.
Developed a hybrid Isaac Lab + PPO manipulation system for Unitree G1, combining scripted motion, residual hand control, curriculum learning, and randomized simulation for bottle pick-and-lift.
Achieved 97.4% integrated success in randomized simulation and transferred the learned policy to a real G1 Inspire Hand, successfully grasping and lifting a loaded PET bottle containing approximately 600 ml.

Physical AIRoboticsComputer VisionReinforcement LearningIsaac Lab

WorldQuant

Apr. 2024 - Jul. 2025

Remote

Found and submitted trading alphas, mathematical models that seek to predict future price movements of financial instruments, with performance evaluated in real-world stock markets.
Developed optimization algorithms using the WorldQuant API and stock market data to automate the mining of trading alphas.

Quant ResearchOptimizationFinancial Data

Artificial Intelligence Lab, National Taiwan University

Jun. 2024 - Dec. 2024

Taipei, Taiwan

Co-developed trustworthy AI recommendation systems with PhD students for BankTaiwan Life Insurance, focusing on minimizing default risk through anti-recommendation algorithms.
Processed large-scale real-world insurance data and engineered robust features to enhance model generalization.
Implemented and evaluated collaborative filtering and anomaly detection models using PyTorch, optimizing for both ranking accuracy and risk mitigation.
Collaborated with stakeholders to officially deploy the model after successful enterprise testing.

Recommender SystemsPyTorchRisk Control

National Institute of Information and Communications Technology (NICT)

Jul. 2023 - Sep. 2023

Tokyo, Japan

Conducted research to enhance the resilience of model-based network intrusion detection systems for IoT devices by investigating advanced adversarial attacks and robust defense strategies.
Engineered and trained diverse NIDS models, including Logistic Regression, kNN, Random Forest, Autoencoder, 1D-CNN, and 2D-CNN, using scikit-learn and TensorFlow on real-world TON_IoT datasets.
Developed a SHAP-guided feature selection algorithm that reduced input features by 65% while maintaining model performance and interpretability.
Investigated adversarial training using SHAP-informed attacks, providing insights for NICT researchers' ongoing studies on adversarial training and defense strategies.

Adversarial MLNIDSSHAP