A T5 model is fine-tuned on the SQuAD dataset to generate accurate questions from input passages using context-question pairs and a text-to-text generation approach, with performance evaluated using BLEU scores.

A multilingual T5 model is fine-tuned to generate summaries from Bengali text using encoder-decoder architectures for cross-lingual understanding and coherent text generation.

A top-down heatmap-based pose estimation model is trained using keypoint annotations in the MMpose framework with an HRNet backbone to accurately detect cattle anatomical landmarks.

Predictive models are trained on structured tabular clinical data to detect heart disease risk using classical machine learning algorithms.