Meta | Senior Machine Learning Engineer (2022 – Present)
Generative Audio Restoration: Architecting production-level GAN models to restore and enhance speech quality across all video content in the Meta Family of Apps (Facebook, Instagram, WhatsApp).
Massive-Scale Quality Assessment: Developed on-device DL pipelines for real-time audio quality metrics (NS/PLC). Partnered with SWE teams to deploy these models to 3 Billion Daily Active Users.
Next-Gen Flow Matching: Leading research into Flow Matching (FM) architectures to achieve high-fidelity audio enhancement with optimized, real-time inference on mobile hardware.
Samsung Research America | ML Research Engineer (2019 – 2022)
Wake Word Ecosystem: Developed production DL models for the Galaxy "Always-On" trigger word pipeline, focusing on low-power constraints and high-noise robustness.
Advanced Model Compression: Utilized Knowledge Distillation and Self-Supervised Learning (SSL) to compress model footprints for wearables and mobile devices without sacrificing accuracy.
Technical Leadership: Guided the Data team in automating global data augmentation and metrics logging, accelerating the deployment of new trigger words.
Graduate Research: Insect Robot Vision & Control (UW AIR Lab)
Developed the hardware and control algorithms for the UW Robofly, a 75 mg flapping-wing robot.
Hardware Design: Custom-built an ultra-low-weight (26 mg) onboard camera system to provide visual feedback.
Onboard Intelligence: Trained a resource-constrained CNN to perform real-time image classification (Predator vs. Prey), enabling autonomous maneuvers based on visual input.
Published at Biorob 2018.
Samsung Research America | Research Intern (Summer 2018)
App Graph Analysis: Leveraged unsupervised clustering and supervised classification to label paths in application code graph data.
Sequence Modeling: Implemented Recurrent Neural Networks (RNNs) for sequence modeling to validate application behavior paths.
Mitacs Globalink Research Intern | University of Victoria (Summer 2015)
Haptic Rendering: Developed a haptic user interface for rendering rigid body dynamics and kinematics using the Novint Falcon Haptic device.
Awarded the prestigious Mitacs Globalink Fellowship.
SMARTMOS (2025): Developed a framework for modeling subjective audio quality evaluation, aligning DL metrics with human auditory perception for real-time applications.
State Estimation & Filtering: Compared Kalman and Extended Kalman Filters (EKF) for state estimation of an inverted pendulum, proving EKF superiority for non-linear dynamics.
Dynamic Homography Estimation: Built a system to track Harris corner points using optical flow and used logistic regression to differentiate between moving objects and stationary backgrounds.
Bio-Inspired Maneuver Simulation: Modeled and simulated the complex food-retrieving behavior of the Saharan Silver Desert ant in MATLAB.
Optimal Control Design: Designed LQR estimators and controllers for the 2-D flight control of insect-scale robots, testing for controllability and observability.
Patents (2024-2025): 3 granted/published U.S. patents in the areas of accent-agnostic wake word detection and wake-word-less command fulfillment.
e-Yantra Robotics Competition: Developed a cargo alignment robot using the Firebird V platform; conducted by IIT-Bombay.
Green Aeromodelling & Aresto Momentum: Awarded 1st Place in technical design competitions at NIT Trichy.
Image Processing & Surface Computing: Certified in advanced image processing techniques through EI Systems.