A two-stage deep learning pipeline classifies Indian sign language video clips into English words using a fine-tuned VideoMAE model and translates them into Hindi, Telugu, and Bengali via the NLLB-200 multilingual model. The system achieves 99% training and 78% validation accuracy on a 13-class, 197-clips dataset with uniform 16-frame clips at 22-224 resolution, and includes a Streamlit demo for user-uploaded videos with per-class analysis and failure mode identification.