Selected work · Archive
A collection of open-source, research, and earlier projects beyond the featured case studies.
University of California, Los Angeles · Jul 2020 – Jan 2023
Diagnosed skin-tone bias in medical remote photoplethysmography and developed an adversarial multimodal fusion algorithm combining 77 GHz radar and RGB video, improving pulse estimation accuracy across diverse skin tones by 75%. Published at SIGGRAPH 2022.
University of California, Los Angeles · Jan 2022 – Jun 2022
RGB camera face videos and finger pulse oximeter ground-truth PPG signals are not hardware-triggered, creating temporal lag between frames and ground truth. Proposed a novel shift-robust regression loss function that learns directly from misaligned signals, decreasing pulse prediction error by 40%.
University of California, Los Angeles · Jan 2022 – Mar 2022
Explored deep learning architectures—including CNNs, LSTMs, RNNs, VAEs, Transformers, and attention mechanisms—for multi-class classification of EEG motor imagery signals, comparing single-subject versus full-dataset training and different signal durations.
University of California, Los Angeles · Apr 2022 – Jun 2022
Studied indoor tracking for gaze estimation using ORB-SLAM on a custom-built wearable eye-gaze setup. Extracted pupil parameters with a modified U-Net architecture and fused IMU and fisheye camera ORB-SLAM pose estimates to calculate user-to-gaze-point distance.
University of California, Los Angeles · Jan 2022 – Mar 2022
Modeled and analyzed the rapid growth of top-end wealth inequality over the past forty years using Itô’s calculus, Brownian motion, and stochastic differential equations, examining deviations driven by skill-price changes and superstar entrepreneur dynamics.
VidMob · Jun 2022 – Sep 2022
Designed computer vision models for object detection and image scoring in PyTorch, lifting ad engagement classification accuracy by 25% and powering automated recommendation pipelines.
Indian Institute of Science (IISc) · Jan 2019 – Jun 2019
Cleaned and rearranged the ABIDE dataset based on correlation between brain regions, and developed deep neural networks to analyze fMRI scans for early Autism detection, achieving 80% accuracy on cleaned neuroimaging data.
PES University · Jan 2019 – May 2019
Engineered a Raspberry Pi robotic crane that segregates waste into biodegradable, non-biodegradable, and electronic waste with 95.18% accuracy, comparing PCA, LDA, and Convolutional Neural Network feature extraction with transfer learning.
PES University · 2018
Estimated human heart rate from 10-to-30-second facial videos by tracking subtle ballistocardiographic head movements with Haar Cascade detection and signal processing, achieving an error margin of 3 to 4 heartbeats.
PES University · 2018
Built an embedded voice-assistant hardware prototype on Raspberry Pi integrating custom PCBs, sensors, and multi-protocol communication for interactive voice control.