Gao Han

Designed and implemented a hierarchical deep reinforcement learning framework (PPO + SAC) with GNN-enhanced path encoding to optimize multi-path DeFi token swaps. Achieved an 84% win rate (95% confidence) against baseline strategies, reducing slippage by ~18% while ensuring stable returns under varying liquidity conditions. Demonstrated 30–40% faster convergence (fewer training steps to stability) and robustness on unseen token swap graphs. Built a full experimentation pipeline with Optuna-based hyperparameter tuning, TensorBoard visualization, and a custom Uniswap V3 simulator to benchmark against Dijkstra (Pathfinder) and brute-force baselines.

Led smart car network security assessments and actively participated in penetration testing of Zeekr models, identifying and documenting 10+ high- and medium-severity vulnerabilities within the testing scope. Utilized Wi-Fi network cards and security testing toolchains to uncover ARM architecture vulnerabilities, contributing to remediation of system-level security risks and enhancing vehicle communication resilience. Mastered key vehicle communication protocols and proactively collaborated with senior engineers to align findings with industry security standards.

Led smart car network security assessments and penetration testing, identifying 10+ high- and medium-severity vulnerabilities. Used security testing tools to uncover ARM vulnerabilities and collaborated on remediation to enhance vehicle communication resilience.