Nov 2022 – May 2023 · 6 mos · Remote

• Part of the founding team of this early-stage startup (formed as a spin-off of Uplara), that works in the Furniture e-commerce space.
• Productised ML algorithms for automated 3D textured reconstruction of furniture.
• Improving the accuracy and saleability of product descriptions via our proprietary taxonomy and ML algorithms.
• Implemented automated contact form answering, social media scraping and market out-reach strategies.
• Played an active role in hiring and marketing.

Sep 2022 – Nov 2022 · 2 mos

• Developed the proprietary sleeve, collar and clothing descriptors (based on pivot ratios, slopes, focus masks etc) which enabled the control of different lengths/depths, starting/ending points and curvatures.
• Extended our algorithm to do simultaneous tryon of tops and bottoms, with different overlapping configurations - tucked-in, overhanging, crop-top etc.
• Worked on several image smoothing methods including our novel probability based combination to enhance the overall quality/blending of the final image.
• Developed a beta version of our shoe tryon algorithm - focused on frontal view tryon.

Aug 2020 – Sep 2022 · 2 yrs 1 mo

• General**
• Was a key part of the early team of scientists/engineers from MIT, CMU, Stanford, IIT-B and IIIT-H at this early stage seed funded startup based out of Silicon Valley.
• Apart from my primary role of developing machine learning algorithms, played a pivotal role in developing processes and systems for team communication, feedback-mechanisms, data curation.
• Was an active part in business decisions - company positioning, customer acquisition strategies, business development strategies, hiring etc.
• Machine Learning**
• Developed multiple machine learning algorithms to perform virtual 2D tryon of several categories - shirts of different sleeve lengths and collars, shorts/pants, dresses with different starting points and collars, crop-tops and shoes.
• In doing so, was responsible for creating superior light-weight models capable of on-device inference for real-world human segmentation, 2d pose estimation, clothing segmentation, image warping, cloth inpainting, skin inpainting and expected segmentation of tryon.
• For attaining superior state-of-the-art results across these tasks, broke down the pipeline into easier sub-problems and developed novel loss functions. Additionally, explored numerous SOTA models such as lightweight/mobile models, GANs, U-Nets, Image Transformers etc. across literature and adapted them for our tasks.

Jun 2019 – Mar 2020 · 9 mos · On-site

• During this effort I worked with Prof. Leonidas Guibas (Stanford), Prof. Srinath Sridhar (Brown Univ) and several post-docs/Ph.Ds. at Stanford on modelling real world dynamics from images.
• The problem I worked on was as follows: Given a video of an object impacted by a random force, our focus was to build an interpretable representation by which we can perform system identification i.e, predict the appearance (pose, shape), intrinsic (mass, volume etc), extrinsic (position, velocity) and material properties of the object as well as exploit temporal dependency.
• As a part of this effort, I was able to create our own synthetic physics dataset and develop multiple models to predict the pose and shape in the normalised object co-ordinate (NOCS) space.

Apr 2017 – Jun 2019 · 2 yrs 2 mos · Center for Visual Information Technology (CVIT)

• As a part of my Research Assitantship and Masters, I explored multiple problems in 3D Computer Vision - 3D Non-rigid body modelling, Exploiting Dense Correspondences, 3D Human Reconstruction, 3D Action Recognition, 2D/3D Pose Estimation as well as Brain Graph Analysis.
• During this period, I successfully completed by Masters by Research on 3D Monocular Human Body Reconstruction with publications in BMVC, ICCV-W and ACCV.
• My research was sponsored by Qualcomm's Innovation Fellowship (QIF), which was felicitated by the Dept. and Science and Technology Head, Prof. Ashutosh Sharma.

Aug 2014 – Dec 2014 · 4 mos · Hyderabad Area, India

• Worked on 'HelloPsych'​, which is an online Video Psychiatric Consultancy website. It was written using the Python based 'Django' platform. I was part of a team of 4 people who worked on it.