At TensorMaxwell Corporation, I developed a PyTorch reinforcement learning pipeline that trained a simulated robotic arm to grasp and stack objects using Soft Actor-Critic in a MuJoCo/robosuite environment. I built a memory buffer for 10M+ past experiences and used TensorBoard to track training.
In my projects, I improved customer churn recall from 47% to 74% with SMOTE and reached 0.83 ROC-AUC with XGBoost. I also built a content-based movie recommendation engine and used K-Means to identify customer segments.
