Best Master's Thesis award by the Department of Electrical Engineering, IIT Bombay 2022.
Finalist in the KNIGHT Challenge at the IEEE 19th International Symposium on Biomedical Imaging (ISBI) conference.
Finalist, Nokia White Paper Contest, Phase Shift - BMSCE Tech Symposium.
Academic Excellence Award for ranking first in school, 10th ICSE Board exam 2012.
Work Experience
Wisconsin Reading Center, UW–Madison, WI, USA
Designation : Data Science Researcher (Aug’25 – present)
Developed a weakly supervised Swin Transformer pipeline for ultra-widefield retinal image analysis on a distributed multi-GPU cluster, achieving 76% detection of severe diabetic retinopathy at high specificity and enabling screening with limited labeled data.
Improved recall by 11.9% by designing and deploying ensemble models via combining EfficientNet and attention-based architectures.
University of Wisconsin-Madison, WI, USA
Designation : Graduate Research Assistant (Jan’24 – Dec’24)
Applied feature engineering and dimensionality reduction (MRMR, RFECV) on DCE-MRI radiomic features to enhance model interpretability and optimize high-dimensional longitudinal imaging data.
Trained and evaluated ML classification models (SVM, Random Forest, XGBoost) to predict pathological complete response (pCR) in breast cancer, enabling precision medicine insights.
Nference, Bangalore, India
Designation : Data Scientist (Jan’23 – Aug’23)
Engineered YOLOv5-based object detection model for semantic segmentation of whole slide images (WSIs), identifying key kidney structures (glomeruli, tubules, vessels) with a mean average precision (mAP) of 0.7 for automated kidney disease diagnosis.
Developed a DenseNet-inspired encoder-decoder network to enhance tissue localization on digitized IHC and H&E-stained glass slide images at 1x resolution, achieving an AUC of 0.95 and outperforming the U-Net architecture.
Worked on adapting a 92.6% smaller transformer-based multi-lingual encoder model to lower latency and training time while maintaining original performance of 2.3+B parameter BERT-based AlexaTM model.
Designed and refined a task classification pipeline to route queries across 4+ Alexa platforms, enabling multimodal interactions through improved transformer-based intent recognition for downstream applications.
Conducted A/B testing and statistical analysis on multilingual traffic, evaluating model performance through precision, recall, F1-score, and latency metrics to validate the efficiency of the new transformer architecture.
Visual Information Processing Lab, Indian Institute of Science (IISc) Bangalore
Designation : Research Assistant (July’18 – May’19)
Improved correlation between the neural network (NN) based objective quality assessment algorithm and human subjective scores by 12% by incorporating subject viewing directions to analyze user visual discomfort with virtual reality (VR) videos.
Trained MLP-based regression neural networks to predict of head motion trajectory using the video frames and previous viewing direction as the input to enable selective video frame optimization of future frames.
Internships
Honeywell Aerospace, Bangalore
Designation : Intern (Jan’18 – July’18)
Developed proof of concept of Trustzone creation and API development for ARM Cortex‑A processors.
Signal Processing, Interpretation and REpresentation (SPIRE) Lab, IISc Bangalore, India
Designation : Research Intern (June’16 – Aug’16)
Implemented EM Algorithm in C to optimize an analytically intractable likelihood function, with an aim of reducing latency.
Implemented Distributed EM algorithm on MATLAB, employed when data is generated at several nodes.