• Leading NIH grant project with Oxford University and collaborating with an interdisciplinary team of Pathologists, Biologists, Data scientists regarding data, annotation, validation and insights

• Distinguished disease type using pre-training with contrastive learning, graph transformers and Vision Transformer segmentation models at patch, interconnected patch and pathology level; currently preparing manuscript for publication

• Developed xAI techniques for visualizing attentions across entire whole slide image in QuPath; trained GAN models to remove marker stains and applied alignment in whole slide images

• Integrated and LoRA finetuned QwenVL, LLaVA LLM into a multi-modal pipeline for automated analysis of crop based attentions heatmaps explaining the distribution, improving inference speed and accuracy

• Designed and trained a closed loop Reinforcement learning based PPO agent for a thinking microscope robot to allow it to tune itself for optimal power and duration to maximize cell lives in wells

• Support grant efforts with preliminary analyses and proposal writing based on literature review

• Developed scalable, automated deep learning pipeline using multi-GPU nodes for segmenting and classifying neurodegenerative pathologies in gigapixel whole slide images

• Deployed trained model for detecting plaques in AD histopathological tissue images in RunPod cloud

• Created inter-rater agreement tool within QuPath using Java Groovy to facilitate collecting annotations from various pathologists with the aim of generating consensus training dataset

• Support grant efforts with preliminary analyses and proposal writing based on literature review

• Built generalized attention-based training framework using pretrained foundation models to classify whole slide images for abnormal tissue detection and track the movement of neurons in microscopic images

• Training foundational model for classifying & segmenting neuronal vs glial cells from microscopic images

Internal product:

• Created a scalable end to end resume-parser for automated resume screening and matching relevant job roles accelerating the hiring process

• Trained YOLO object detection model for detecting paragraph, heading, photograph in resume; BERT classifier for headers; extracted industry-skill years of experience, title matrix using text algorithms

and libraries like Regex, BERT, Simstring etc

• Extracted features such as section, skills from Job description and created weighted ranking for role and resume match; deployed in AWS EC2 Nginx; used Flask/Streamlit for API service

UAE based Client with Leisure & Entertainment business:

• Identified high lifetime valued customers with 45% recall improvement over baseline using LSTM based time series regression approach; experimented with Random Forest, LGBM, XGBoost models

• Improved 8% revenue gain and increased package incidence of client’s market by determining optimal pricing of packages using clustering with Entropy based Discretization technique and Simulated Annealing approach

• Evaluated performance across concepts and screens by sites and markets; built an AWS alert system to recommend relevant business action

• Performed sentiment analysis of NPS customer survey data of malls using NLTK tools

US based client with e-com retail business:

• Segmented buying behavior patterns for similar items using order transactions (used K-mean clustering)

• Identified potential bundle candidates and likelihood of buying together using Apriori Algorithm

• Generated similarity ranking of recommended bundles based on product images using ANNOY indexing and CNN - VGG network and product metadata

US based client with transportation business:

• Used LSH algorithmic approach to find the control customer highly similar to each test customer

• Performed Statistical tests and analyzed performance of pre and post enrollment

Advanced Programming Lab, Introduction to Programming Lab

As a researcher, I worked on my M.Tech thesis "Performance Evaluation of SP Methods for Simulation Optimisation & Policy Learning" under the guidance of Prof. Prashanth L.A. Overview of my work:

• Evaluated and compared performances of stochastic optimisation algorithms using Simultaneous Perturbation methods on multimodal functions with increasing dimensions

• Developed policy gradient using Simultaneous Perturbation techniques to train Neural Network for optimizing parameterized policy in Reinforcement Learning problems

• Obtained best parameter settings for classic control environments with discrete and continuous action space and compared performance with REINFORCE algorithm