Jan 2021 – Jun 2021 · 5 mos · Chicago, Illinois, United States

• - Worked with high volume data, developing efficient and optimized data collection and processing pipelines for claim estimation machine Learning models, extensively using PySpark, Pandas.

Jun 2020 – Aug 2020 · 2 mos · Seattle, Washington, United States

• As part of Zestimate AI team, worked on incorporating comparable homes(comps) in the Rent Zestimate pipeline to be displayed on Rental Manager and Home Display Page to customers. Successfully generated comps nationwide for over​ 100 million​ homes.
• Productionalized a ML oriented similarity based comps generation method in ​Python, ​deploying workflow on ​Kubeflow ​pipeline and storing in​ AWS S3.
• Achieved​ National Coverage of ~98%​ and carried out memory analysis, tuning and evaluation.

Oct 2019 – Jun 2020 · 8 mos · Greater Los Angeles Area

• Developed a novel end to end ​spatial ensemble framework for ​land cover mapping applications using Deep Learning and Computer Vision techniques, catering to input and class distribution shifts in satellite imagery, under supervision of ​Dr. Bistra Dilkina​ and a PhD student.
• Achieved an average ​accuracy gain of 3.5% over baseline land cover models. Worked with frameworks & libraries like ​rasterio​(to handle geospatial data), ​Pytorch​ and ​Matplotlib​. This work was accepted at AGU Meeting'20

Aug 2017 – Jul 2018 · 11 mos · New Delhi Area, India

• Center For Artificial Intelligence (Autonomous Vehicle Lab)
• Adviser : Dr. Saket Anand
• Led development of the Lane Detection and Tracking module as part of perception team, catering to Indian road scenarios.
• Proposed a novel robust and real-time framework leveraging Deep CNN, Extended Hough Transform and Kalman filter techniques achieving 25% gain in accuracy(to 95.2%) and 2x gain in average running speed(to 38 FPS) compared to previous model.
• Implemented module in Python and C++ with various frameworks & libraries like PyTorch, Numpy, OpenCV and Matplotlib, further integrating with car system using ROS.
• Contributed to new Indian road dataset with lane annotations and carried out extensive analysis and processing on it.

Feb 2017 – Apr 2017 · 2 mos · Remote

• Developed pipeline for car segmentation and detection for automated driving test assessment application attaining IoU(Intersection over Union) of 86.3% and 88.6% for segmentation and detection respectively .
• Implemented and analyzed various CNN based architectures like DeepMask,FCNs and Fast R-CNN with respect to speed, accuracy and scalability of project.
• Implemented pipeline with help of PyTorch,TensorFlow and Caffe Deep Learning frameworks, deployed on AWS for training and testing.

Sep 2016 – Dec 2016 · 3 mos

• DRI(Directly Responsible Indivisual- one of 2 leaders) for "Deep Residual Learning for Image Recognition" talk.
• Student Researcher as a part of the Stanford Scholars Program, focused on making world's research more accessible, by collaboratively creating short 10 mins talks on influential papers in many areas of computer science including HCI, crowdsourcing, Computer Vision and Machine Learning, under the guidance of Dr Rajan Vaish (Post-Doctorate Researcher at Stanford University)