Formulated a novel subjective table-to-text generation task, built a new dataset, and outperformed LLMs with a fine-tuned model achieving the highest metric.

Developed a state-of-the-art seq-to-seq model using T5-base to generate coherent text from RDF triples on WebNLG, with comprehensive ablation and evaluation.

Performed argument generation on CMV dataset using multi-task prompting on T5 model. The model was trained to optimize all the losses of different objectives.

Implemented a model to identify correct meaning of an ambiguous word through picture input. Created a novel sample dataset to test the results.

Building a website to find various homes or apartments for buying, renting or leasing based on different demographics, features and facilities.

Developed a lightweight container from scratch, and analyzed its live migration downtime through experiments.

Built a webserver with multiple threads, capable of handling many clients at the same time over TCP sockets. HTTP requests are served successfully and also the performance bottlenecks are optimized.

Developed two seperate models for POS Tagging on the universal tagset using Encoder-Decoder architecture and Feed-forward Neural Network.

Implemented functional properties of a linux shell. Background process, foreground process, forking, reaping, process memory management and signal handling have been implemented along with efficient testing.

Analysed various social, economic and dynamic challenges faced by a startup system. Modeled them and simulated to find right situation for a startup growth.

A dataset of personal life-based answers of tech people is used to predict mental health. Various models are used and the best of them are mentioned.

Our proposed system will automate processes with better accuracy, integrate barcode-based patient and inventory tracking, ensure safety and social distancing, offer a user-friendly and expandable interface, with minor added costs for cloud and technology integration.

We proposed peptide vaccines and developed a 2D Polar plot method using biochemical properties of amino acids to compute a characteristic qR value for nucleotide and protein sequences.

A simple model using LSTM to fit a curve in stock prices of previous recordings.

Random walk is a popular model of statistics which represents the possible path distance with randomisation with a given time quanta.