Jul 2021 – Aug 2022 · 1 yr 1 mo

May 2019 – Jul 2021 · 2 yrs 2 mos

• Sensor fusion of full sensor suite (cameras, radars, stereo, imu and ultrasonics) to create a local object map (C++).

May 2018 – Dec 2018 · 7 mos · Atlanta, Georgia

• ▪ Developed a custom node for multi object lead vehicle tracking using sensor fusion of vision and radar data employing Kalman filters and data association techniques.
• ▪ Trained YOLO network architecture (in Tensorflow) for object detection improving real time speed (45 FPS) over existing architectures such as SqueezeDet and developed a wrapper over object detection using KLT tracker to remove false negatives.
• ▪ Currently developing an offline auto annotation software involving amalgamation of instance level heavyweight segmentation (Mask R-CNN) and object detection algorithms (Yolo V3) for extremely accurate annotation.

Dec 2017 – Aug 2018 · 8 mos · Greater Atlanta Area

• Adversarial boosting algorithm that prevents external attack on machine learning systems (Under Prof. Sebastian Pokutta).
• ▪ Used classifier evasion technique to generate adversarial examples for tree-based classifiers such as Random Forest for MNIST handwritten digit recognizer.
• ▪ Designed a novel adversarial boosting algorithm by training it on adversarial examples that increases the robustness of classifier.
• ▪ Accuracy (on adversarial examples) shot up from 67.5% to 80% after adversarial boosting for MNIST dataset.
• ▪ Employed parallel processing to train the system on 32-core machine with total memory of 128 GB.

Jan 2017 – Aug 2017 · 7 mos · New Delhi Area, India

• Time Series (Energy Data) analysis and forecasting using python data analytics stack.
• Responsible for designing an anomaly detection engine that generates alert in case of possible gas leakage of air conditioning system for the client.
• Used ensemble of supervised (support vector machines) and unsupervised methods (clustering methods including density based clustering such as DBSCAN) for gas leakage signature detection on the energy data (power, temperature) of one year duration coming from hundreds of energy meters after data cleaning and statistical feature extraction.
• Mining complaints data to come up with a business metric called gas leakage detection time for evaluation of the system.
• System was able to dramatically reduce the gas leakage detection time (from ~13 days to ~5 days) validating using test complaints data.

Jun 2014 – Aug 2016 · 2 yrs 2 mos · NSIT, Dwarka, New Delhi

• • Worked on different Control Strategies and System Identification techniques.
• • Gained hands on experience of different Hardware modules such MS15 DC motor, XPO-FL Process Control Trainer and Quanser Omni Bundle Robotic Arm and their data acquisition using NI based PCI-6221 DAQ card.
• • Gained working proficiency in coding environments like MATLAB and LabVIEW during project work at Advanced Process Control Lab.
• • Designing and modeled a real Time Control Strategy for three link Quanser Omni Bundle Robotic Arm Manipulator.

Dec 2012 – Feb 2013 · 2 mos · New Delhi

• Learnt basics of embedded systems and electronics. Played with LEDs, sensors and microcontrollers.