Building a real-time ASL word recognition pipeline on the WLASL100 dataset (100 signs, 2,000+ video samples) using MediaPipe Hands for landmark extraction. Extracting and normalizing 21 hand landmarks (63 features) per frame relative to the wrist to ensure position and scale invariance. Benchmarking classifiers (Random Forest, SVM, MLP) on accuracy, F1-score, and real-time inference speed; targeting live webcam deployment with temporal smoothing.