Jun 2023 – Oct 2023 · 4 mos · Delhi, India · On-site

• Working with multimodal learning and 3D GAN’s for speech and head avatar generation and animation

Feb 2022 – Mar 2023 · 1 yr 1 mo · Menlo Park, California, United States

• Worked with the computer vision team to provide 3D vision and graphic rendering solutions for developing and shipping the flagship 3D tablet Leia LumePad 2.
• Improved face and head tracking by trajectory calculation using Kalman-based approaches and known human movement patterns for AR applications.
• Created ML framework for Mono depth estimation and Video layout detection for devices from the ground up using Pytorch, and OpenCV.
• Deployed a simulation system for dataset synthesis built on Unity and Unreal Engine for 3D depth capture deployed on AWS machines for a cloud-based system.
• Deployed computer vision APIs in Amazon Sagemaker for cloud-based API’s.
• Created the OpenGL-based rendering pipeline for deploying synthesized stereo images to displays. Reduced rendering time by up to 20%
• Model optimization for DSP deployment using SNPE and quantization.

Aug 2020 – Feb 2022 · 1 yr 6 mos · San Diego, California, United States

• Component Detection
• Developed a fully automated system for component detection on PCB FOV Images using SFM and Faster-RCNN based object detection in Tensorflow.
• Model optimized in ONNX and Libtorch to meet required speed and efficiency requirements.
• Achieved state-of-the-art performance in both detection and speed with a 96.2% capture rate.
• Anomaly detection and segmentation
• Designed a Kernel Density Estimation based fully automated pipeline for detection and segmentation of anomalies on circuit boards.
• Outperformed deep learning-based approaches in per-pixel accuracy and generalization with very few examples of anomalies available.
• Model implemented with Thread Building Block(IntelTM TBB ) for
• threading-based parallalization and CPU efficient optimization reducing run time by 35%

Sep 2019 – Feb 2022 · 2 yrs 5 mos · La Jolla

• Working with Prof. Gal Mishne on Calcium imaging registration methods
• Multi session affine image registration using graph spectral features.
• Current pipeline brings down the work time for neuroscientists from hours to
• minutes

Mar 2019 – Jun 2019 · 3 mos · La Jolla

• Worked with Dr. David Kriegman on order invariant SFM reconstruction for corals. Tackled the problem of order invariance as a set invariance problem. Employed the concept of deep sets for the same.

Jun 2019 – Aug 2020 · 1 yr 2 mos · San Diego, California

• 1)Worked with the core computer vision team to develop a component detection system
• 2)Developed state of the art system for component detection on PCB boards using a combination of conventional computer vision and deep learning
• 3) System to be deployed in next iteration of scanners.

May 2017 – Jul 2017 · 2 mos

• Worked with the Bing team to improve ad recommendations. Used deep visual tokens to optimize results for a search query and rank products for ad recommendation. Generate textual token from an image using a visual attention-based image captioning model using an LSTM. Used the tokens to describe the focus and content of the image.

Nov 2016 – Mar 2017 · 4 mos · Hyderabad Area, India

• We created an original game, by combining the concept of tower defence and typing tutor. We handcrafted towers and models in blender. The game was made available on Windows play store as the part of Hackathon.
• Our team was among the top 10 in the national finals.

May 2016 – Jul 2018 · 2 yrs 2 mos · Hyderabad Area, India

• Advisor: Professor Jayanthi Sivaswamy
• January 2018- May 2018: Developed a method for retinal abnormality detection using a network of CNN's and geometric transformations.
• May - December 2017 : Developed a novel method for automated segmentation and detection of retinal fluid in OCT scans for diagnosis of diabetic retinopathy. This work employed generalized motion pattern and CNN for augmenting the scans and improving segmentation.
• January - May 2017: Developed a super pixelation and random forest based method for detection and segmentation of tumors in liver CT scans for LITS challenge at ISBI 2017.
• 2016-2017: Developed a method for super ressolution of surveillance footage using an LBP characteristic model based approach.

Aug 2015 – Nov 2015 · 3 mos · Hyderabad, Telangana, India

• Developed an Android app for mobile payments starting from scratch as a part of a larger integrated environment.