Experience

Developing NLP and Computer vision-based solution for faster reviewing of contracts and generating insights from contracts. Developed a Generative LLM based solution (one of the first at Swiss Re) to process the contracts faster reducing reviewing time from anywhere between 1 week-1 month to 3-5 minutes. Handling lots of unstructured data to find patterns and generate helpful understanding for the users  

Developed pragmatic deep learning-based solutions for Table Detection and Structure Recognition using ResNeXt backboned Faster R-CNN network. Worked on novel IoU based metric and loss function for object detection problem. F1 score as high as 0.96 achieved for cross-testing set-up creating a new benchmark in this domain and beating the previous SOTA by Microsoft Research 

Worked on the automated sign language detection project for helping the people with hearing disabilities. Also exercised data extraction skills using web automation (selenium) and other codec tools

Architectured an Unsupervised Compound Domain Adaptation set-up for Face-Anti Spoofing. We proposed a memory augmentation method for adapting the source model (ResNet-50) to target domain in a domain aware manner. The adaptation process is further improved by using curriculum learning and the domain agnostic source network training approaches. 

Our experiments demonstrate superiority over state-of-the-art method with HTER 20% and AUC 85.8% (accepted in IEEE International Conference on Automatic Face and Gesture Recognition 2021) [LINK]

Teaching Assistant for "Machine Learning for Natural Language Processing 1" in Fall Semester 2020. Responsible for taking tutorial lectures, clarifying Q&A on the online forum, prepare and evaluate exercises from basic (Logistic Regression for Text Classification) to advance level (BERT using Huggingface library), share useful study materials and provide technical assistance.

I invested the summer of 2018 doing research in Carleton University Biomedical Informatics Co-laboratory (CUBIC) at Carleton University, Ottawa, Canada where we applied machine learning, computer vision and data science to solve problems in biomedical informatics. I worked under the guidance of Dr. James Green along with the industrial collaboration with Clearwater Clinical limited. The project involved the classification of ShoeBOX audiograms. Clearwater Clinical has developed a ShoeBOX audiometry application to conduct automated diagnostic hearing testing outside of a sound booth for non-Audiologists. 

My first rendezvous with university research was in the summers of 2017 when I visited Linköping University, Sweden as a Research Intern under Prof. Mario Garrido for about 10 weeks. During that internship, I learnt about deep neural networks like CNN and RNN. The objective of the research was to understand the feasibility of implementing Convolutional Neural Networks for Numerical Detection on the Field Programmable Gate Array (FPGA) whilst reducing power consumption.

Under the guidance of Prof. Satyabrata Jit, I designed and simulated an Advanced MOSFET device, SD-MAGFET using ATLAS programming language and implemented circuit of SD-MAGFET, which consists of a set of off-chip balancing resistors, and improve sensitivity while achieving optimal results on various graphs.