Powering Real Impact with Reliable Code

Comprehensive comparative study evaluating Neural Machine Translation techniques using RNN-based LSTM, GRU models, and Transformer architecture. Trained on over 220K sentence pairs and evaluated using BLEU and ROUGE metrics. Results demonstrated that Transformers significantly outperformed RNN variants in translation fluency and contextual accuracy.

Novel Transformer-based pipeline for generating concise textual summaries from videos with frequent scene transitions. System identifies keyframes using clustering, extracts features via InceptionV3, and generates captions using Transformer encoder-decoder. Achieved BLEU-4 score of 52.17 and ROUGE-F1 score of 40.21.

Innovative semi-supervised learning approach using modified MoCoGAN architecture for video classification with minimal labeled data. Discriminator performs dual tasks of distinguishing real/fake videos and classifying across six categories. Achieved over 62% accuracy with 50% labeled data, outperforming traditional supervised baselines.

Comprehensive dual-user web application designed for early Autism Spectrum Disorder diagnosis and personalized learning support. Utilizes AdaBoost and SVM models to predict autism severity levels. Features interactive emotion, sound, and audio-based quizzes tailored to individual severity levels with comprehensive progress tracking for caregivers.