Stealth

- Pioneered hypernetwork and diffusion model architectures to generate photorealistic virtual garments that adapt seamlessly to diverse body types. - Led a remote team of 10 engineers to build scalable AI infrastructure capable of training models on massive datasets of fashion items (millions of SKUs).- Implemented body mapping algorithms using advanced computer vision techniques to ensure garment fit accuracy across a wide range of body shapes.

Led cross functional team of software developers and research scientists.Developed novel facial recognition and entity mapping algorithms, increasing performance of facial recognition systems.Led infrastructure and MLOps efforts to increase system bandwidth and latency 5-fold, enabling the system to serve hundreds of thousands of requests seamlessly.Developed unique algorithmic and operational approaches to increase safety against malicious actors, reducing the number of breaches by 71%

Google Brain: Model Interpretability at PAIR.Making AI usable by anyone. Bringing TCAV to the masses.

Google Research and Machine IntelligenceFaceUnlock for Pixel 4

Built a fully functional Deep Learning system for product recommendation from scratch. Built pipeline to get data from multiple sources inside of the Google Cloud platform, transform this data and send it to the machine learning model. Built and designed a Neural Embedding Model that takes advantage of product metadata and other information.Main objective was to recommend products in the Home Depot website.Model was designed after reading several academic papers and implemented using Tensorflow, Google CloudML and BigQuery. Created automated data pipelines in the Google Cloud Platform. Those can easily move credit card data to Citibank users and Data Analysts in Home Depot. This data could then be analyzed for decline and fraud errors, saving the company 5 million USDBuilt tool to diagnose health of multiple services throughout the company, enabling users to easily detect errors in their systems and report useful metrics to leadership. Tool can be integrated with any Google Cloud project. Used Google Cloud Datastore,Google Cloud Pubsub and Java.

- Worked on developing deep learning algorithms that operate on Cardiac MRI images.- Worked with manipulating SSFP and DENSE MRI images. The project's long term goal was to create a AI model that could easily convert between the old SSFP format to the newly created, more information rich, dense scan- Created different benchmarks for motion detection in heart imagery. Also applied multiple different algorithms in order to predict heart tissue motion between frames.- Implemented a neural network capable of computing optical flow based on the academic paper FlowNet: Learning Optical Flow with Convolutional Networks.

Worked on creating machine learning methods to quantify humor and recommend jokes to different users. Some of my work in this project included:- Created metrics to evaluate machine learning models on the datasets. - Used Matrix Factorization to recommend content to users based on their past preferences. - Used K-means clustering to cluster content and understand similarities between jokes-Worked with SQL to filter content based on different features such as user likes, favorites, amount of views, etc.This project was built using the Azure machine learning platform