Hey ! I am a Masters student studying ECE at the University of California San Diego. I work at the intersection of Robotics, Computer Vision and Machine learning.

Prior to joining UCSD, I had been working as Machine Learning Scientist at TomTom and Fractal.ai for more than 4 years, where I worked on building and deploying machine learning and computer vision models at scale

I have also had the fortune of working in robotics and computer vision research with Prof. Madhava Krishna at IIIT Hyderabad and Prof. Michael Yip at UC San Diego

Email  /  Resume  /  Linkedin  /  Twitter  /  Github

profile photo
Industry Experience
Senior Data Scientist | TomTom, Inc.
November 2021 - August 2022

Worked with Map Making Platform team at TomTom, Inc. on building computer vision and machine learning models for aerial perception, automated integration and generation of road geometry using satellite image data for ingestion into real-time navigation unit of TomTom’s ADAS navigation maps .

Tools and Technologies : pytorch, pytorch lightning, Azure ML, Azure Databricks, XGboost, LightGBM, scikit-learn, pandas, numpy
Machine Learning Scientist | Fractal, Inc.
June 2018 - October 2021

Worked with the Image and Video Analytics team at Fractal on the product offering of a platform for image and video understanding using deep learning and computer vision. Also, worked with Fortune 500 clients such as Sky UK , Mars and Procter & Gamble on projects involving forecasting and predictive analytics using machine learning.

Tools and Technologies : python, pytorch, tensorflow, OpenCV, docker, git, fast API, django, Google Cloud Platform (GCP), Airflow, Jenkins, Numpy, Pandas, Scipy, Scikit-learn

Research Experience
Research Asssistant | Advanced Robotics and Controls Lab, UCSD
January 2023 - Present

Currently working with Prof. Michael Yip on robotics manipulation of deformable objects using 3D vision-based forward and invese dynamics for goal conditioned manipulation with appliction in surgical robotics. Also, worked on building SE(3) equivariant and deformation invariant neural representations for tasks such as grasping, manipulating, or interacting with deformable objects

Tools and Technologies : python, pytorch, pytorch lightning, OpenCV
Research Intern | Robotics Research Centre, IIIT Hyderabad

Worked with Prof. Madhava Krishna developing pipeline for online incremental localization and mapping (SLAM) for indoor 3D scenes using implicit representation formulated by neural radiance fields (NeRF) and finding relative pose between NeRF submaps

Tools and Technologies : python, pytorch, pytorch lightning, OpenCV, Open3d, transforms3d, pybullet
Open Source Projects
Controllable subject guided text-to-image generation and editing using diffusion models
Paper / Code

Developed pipeline for personalized text-to-image generation and editing using latent diffusion models (generative AI), incorporating LoRA finetuning for image generation and cross-attention guidance for image editing (pytorch / hugging face diffusers /stable diffusion)

Tools and Technologies : python, pytorch, huggingface transformers, hugging face diffusers, stable diffusion, OpenCV, transformers
Stereo visual SLAM with local bundle adjustment on KITTI dataset
Code / Demo Video

Implemented a Visual SLAM pipeline for 6-DOF camera pose estimation and outdoor scene mapping using techniques in multi-view geometry - RANSAC, feature tracking, 3D reconstruction, pose estimation (PnP algorithm) and bundle adjustment

Tools and Technologies : python, numpy, open3d, openCV, transforms3d, matplotlib, scipy
How to Extract Roads from Satellite Images using Semantic Segmentation?
Paper / Code

Developed framework for extraction of road geometry from satellite images using convolutional neural network (CNN) based semantic segmentation architectures experimenting with data augmentations, schedulers, optimizers and loss functions

Tools and Technologies : python, pytorch, pytorch-segmentation, torchvision, pillow
Particle Filter SLAM for an indoor differential-drive robot
Paper / Code

Created Particle Filter SLAM pipeline using IMU odometry data and LiDAR scans from sensors mounted on the differential drive robot to enable localization and build occupancy grid map of the environment

Tools and Technologies : python, numpy, matplotlib, OpenCV
MedVAE: Generating Chest X-Ray Images using Variational Autoencoders (VAE)
Paper / Code

Programmed variational auto-encoder (VAE) trained on a dataset of chest X-rays from patients diagnosed with pneumonia to generate realistic and representative chest X-ray images for improved diagnostic

Tools and Technologies : python, pytorch, numpy, matplotlib, OpenCV, matplotlib
GAN-Art: Generating Artwork using Deep Convolutional Generative Adversarial Networks (DCGANs)
Paper / Code

Programmed deep convolutional generative adversarial network (DCGAN) trained on 50,000+ images from 10 different artistic styles to generate high-quality and realistic artwork images that closely resemble the characteristics of each artistic style

Tools and Technologies : python, pytorch, numpy, matplotlib, OpenCV, matplotlib
Optimised orientation tracking using Riemann stochastic gradient descent
Paper / Code

Worked on orientation tracking of a rotating body using quartenion kinematics on IMU data, constrained optimization using Riemannian stochastic gradient descent and generating 360-degree spherical panorama of the indoor scene

Tools and Technologies : python, jax, numpy, matplotlib, OpenCV, matplotlib
Object-centric visual and spatial representations on multi-view CLEVR dataset
Doc / Code

Researched on momentum contrastive learning to build view invariant visual and view-dependent spatial object centric embeddings to build scene graph on multi-view CLEVR dataset

Tools and Technologies : python, pytorch, numpy, matplotlib, OpenCV

Template credits