Shreyansh Bhatt

I worked on the in-memory distributed and parallel graph database development. I worked on several aspects of this project.

My Major Contributions:
· Designed and developed an IO module which includes parallel ingestion and various output utilities.
· Developed a distributed in-memory graph processing system (Pregel abstraction) for a graph query engine to create a hybrid query and analytics engine.
· Designed and developed more than five distributed and parallel graph algorithms.
· Identified and implemented features required for processing more than five billion edge graphs for the system.

On this project, our goal was to teach a self-driving car to see. I worked on this in my efforts to obtain Udacity's self-driving car engineer nanodegree. However, all of the project was performed using real-world data.

Real-world Data Used:
· CNN for traffic light identification with 95% accuracy. CNN was also used for behavior cloning to successfully drive a car in a simulator.
· A fully convolutional neural network to identify cars and road in a video with 85% F-score and 10 FPS (Top 50 Lyft perception challenge). FCNN for road identification with 98% accuracy.
· Vehicle tracking and object detection in a video using SVM.

This project involved multimodal data aggregation.

My Major Contributions:
· Implemented Extended Kalman Filter and Unscented Kalman filter to track bicycle using Radar and Lidar data.
· Implemented Particle filter to localize a car using radar, LiDar and landmark data.
· Developed an integrated data project that successfully drives a car on a test track.

I led the computer science portion of this project which aims at identifying the type of correlation between diversity and intelligence in social media.

My Major Contributions:
· Designed and developed a Word2vec-based text summarization and distance measure based diversity computation to improve fantasy sports player selection. It resulted in a publication.
· Designed and developed a multi-view clustering and multi-objective optimization approach to select optimal diverse crowd. It resulted in a publication.
· Developed a random forest-based tweet classification and binomial test based clustering to identify users who were applying different strategies to select diverse individuals.
· Designed and developed knowledge graph-driven coordinated ascent-based community detection and characterization approach that outperforms state of the art community detection approaches by 20%.
· Designed and developed a generative model and a knowledge graph-based approach to select diverse crowd within a domain to improve collective intelligence.

I developed a knowledge graph and a framework which translates the code matching query as a graph traversal query to identify the correct ICD code.

On this project, I worked with a team of three to develop a MySQL back-end to support GraphQL.

My Major Contributions:
· Investigated an efficient joiner to support MySQL queries.
· Designed and developed a framework in Java to support efficient GraphQL queries.

The project aimed to provide insights into political parties. I led the project and worked on almost all aspects of the project starting from client interaction and requirement gathering to final deliverables.

Major Deliverables:
· An automatic visualization update framework to update the results every week.
· NLP and graph algorithm-based techniques to identify possible next followers and major election issues per constituency.

It resulted in more than four news articles.

I developed a wrapper for Neo4j to store data on multiple computing nodes. I partitioned the graph storage and supported the parallel graph traversal.

The project is about leveraging social data in an emergency response.

My Major Contributions:
· Data cleaning and annotation for identifying “seeker” and “supplier” from the social data.
· NLP and a knowledge graph-based approach to identify the location from the Tweet text.