Built an end-to-end CV + NLP pipeline for continuous sign language recognition — from raw video to natural language output in real time. · Extracted pose features using MediaPipe, trained a custom Temporal Convolutional Network (TCN) from scratch for gesture recognition, and fine-tuned a T5 transformer for natural language translation. · Achieved 99.6% recognition accuracy and 81.24 BLEU-1 score on the W-THISL dataset. · Optimised inference to sub-100ms latency via asynchronous frame processing — enabling stable 30 FPS real-time translation. Published in Springer — International Journal of Data Science and Analytics (April 2024).