Junwei Lu

Experience

1. Software Programmer

   Healthetile

   Jun 2025 — Present · US

   Gesture Recognition Model: Developed a multi-class model for smartwatch deployment, achieving 93.47% accuracy in classifying X, O, and Random gestures from 3-axis accelerometer data using Decision Tree modelsData Pipeline Optimization: Implemented random shift augmentation strategy and 3-stage feature selection pipeline to handle class imbalance and optimize 10 statistical features for embedded deploymentSignal Processing & Preprocessing: Designed end-to-end preprocessing chain including magnitude calculation, windowing segmentation with 50% overlap, and 8-bit quantization for microcontroller deploymentData Validation: Identified and resolved data leakage issues in validation methodology, implemented robust train/test splitting, and achieved 2x data augmentation while preserving class relationships in imbalanced dataset