I am a Ph.D. student at the Cylab biometrics center, Carnegie Mellon University where I work with Prof.Marios Savvides. Much of my research is focussed towards developing real time AI solutions for real world problems in the areas of computer vision and deep learning.
I have about 5 years of experience in training and evaluating deep CNN models for various computer vision tasks like Detection, Classification and Segmentation. My Ph.D thesis applies multiple CNN models in combination with other algorithmic sub-modules towards building a real time Out Of Stock detection system for retail shelves. The system was deployed in 550 Walmart stores across the US in partnership with Bossanova Robotics.
Post working with Bossanova Robotics, we have also recently partnered with another visual AI company called Oosto where we work with their team towards building low cost compute efficient ML models for the EDGE through techniques like knowledge distillation, model pruning and network quantization.
Apart from focussing on the research, I have also TA'd for two courses at CMU. I assisted my advisor as his teaching assistant for a graduate level class called 'Pattern Recognition' where I handled homework grading and conducted weekly recitation classes of 1 hour to go over some important concepts for the students in Fall 2021. Another class I TA'd was a capstone class for the undergraduate students where I mentored 6 teams of 3 students each towards the successful completion of their undergraduate capstone projects in Spring 2021.
Before joining CMU, I was a Masters student at the University of Illinois at Chicago where I had the honor and privilege of working with Prof.Milos Zefran. We focussed on building a multi-modal communication interface in the ROS framework to assist the elderly people in their homes with a dialogue act classifier and a haptic glove.
Apart from work, I enjoy good food, like working out and have recently also forayed into online gaming. If you enjoy Pubg and would like to team up, do drop me an email!
- Train deep CNN models with Pytorch for the multilabel classification task.
- Data collection and annotation was performed as there were no readily available datasets for the task.
- Optimize the model with TensorRT for EDGE deployment.
- Research the current state of the art methods for model training and apply them to on going projects.
- Developed a system to detect out-of-stock labels and misplaced products on retail shelves with PyTorch.
- Assisted the software team in writing production-grade software for the system in Python3.
- 6 patents were published as part of this work.
- Programmed industrial robotic manipulators from FANUC for welding and printing on aluminium sheets.