Biography

I work with Uber as a Senior Software Engineer in the Maps ETA Predictions team. I build ML models and supporting systems to provide ETA estimates for Uber's various usecases. Previously, I worked in the Maps Error Detection team to identify anomalies in Maps using ML models. I completed my Masters of Science in Computer Science at Georgia Institute of Technology, Atlanta with a focus in Machine Learning in Natural Language and Computer Vision. I spent my summer of 2018, interning at Uber Maps. Before Tech, I worked at BlueJeans Networks as a Senior Software Developer.

I completed my Bachelors(Hons.) of Engineering in Computer Science and Masters(Hons.) of Science in Mathematics from Birla Institute of Technology & Science, Pilani

Projects

DeepETA: How Uber Predicts Arrival Times Using Deep Learning

 Learn how we built DeepETA to predict arrival times using deep learning

YODO (Inverse Reinforcement Learning)

 This project aims to teach an agent to stack blocks together in a simulated 2D world using Inverse Reinforcement Learning

Ask Me (Question Generating Agent)

 We developed an agent that could learn to ask a question provided an informative sentence based on a Sequence to Sequence Deep Neural Net architecture powered by Global Attention.

Apache Spark Tuning Notes

 Notes on Apache Spark - A quick reference to all things spark

Hidden Markov Models

 Hidden Markov Model is an excellent tool in the analysis of noisy unlabelled data sequences

Paper Summary: Very Deep Convolutional Networks for Text Classification

 Paper Summary of Very Deep Convolutional Networks for Text Classification - Lecun et al

Scene Classification

 Scene Classification with VGG-like Deep Neural Networks

Surface Reconstruction from Point Cloud Data

 Water Tight Surface Reconstruction of 3D Point Cloud Data using the Ball Pivoting Algorithm

Face Detection

 Face Detection based on Dalal & Triggs' method of Histogram of Gradient descriptors for Pedestrian Detection

Automatic image processing using Expectation Maximization with Gaussian Mixture Models

 We train a mixture of Gaussians to represent an image, and perform automatic image segmentation using Gaussian Mixture Models

Scene Recognition

 Scene Recognition pipeline based on Bag of Words as a feature descriptor. We experiment with multiple features and models for the classification task

Raytracer

 OpenGL/C++ based Raytracer program

Guide to Forking Pixyll

 This website was made possible by the great work of John Otander. Pixyll is available for anyone to fork under the MIT license.