CODE

GitHub

Computer GraphicsNeural NetworksSocial Networks

VIPER Scraper

OU Data Analytics Lab | 2019

The VIPER Scraper is an open-source system for scraping and ingesting multi-model data from Twitter and Instagram. The system includes You Only Look Once (YOLO) real-time object detection integration for Twitter, allowing users to analyze image data as it comes off the data stream. This is useful for uncovering relations between image data and natural language or other metadata, all of which is easily accessable with the VIPER system.

An example of a photo, with classifications, from the VIPER system.
An example of a photo marked up by YOLO, pulled off the Twitter stream
The VIPER scraper was developed to support projects at the University of Oklahoma Data Analytics Lab, including the Visual Inspection of Personal Exposed Records (VIPER) project.


Predicting Atomization Energies of Organic Molecules using Deep Neural Networks

Dr Yihan Shao Research Group | 2017

Worked as a part of the Dr. Yihan Shao Research Group from January ‘17 to August ‘17 to develop a deep neural network capable of predicting the atomization energies of organic molecules. Traditionally, the calculation of a molecule’s atomization energy is computationally expensive, slowing down the work of researchers. However, by using DNNs it is possible accurately predict atomization energies, thereby freeing computational resources and improving the workflow of the research team.

The neural network was developed using Python and Google Tensorflow and is accurate within a 2% error over the chosen set of organic molecules. Research problems include mapping coordinate representations of molecules into data formats suitable for program input, improving the structure and performance of the network, and effectively handling large data sets for training.

Take a look at the source code here.


Hackathons

HackUTD logo

HackUTD

AskUTD is an Alexa skill for UTD students providing live parking information, crisis hotline numbers, contact information for various offices, and fun facts about UTD.

AskUTD was developed with Node.js, AWS Lambda, and the Alexa Skills Kit.

This hack won first prize for the OIT-AWS challenge!

Hacklahoma logo

Hacklahoma

GeoTwit is a data analysis workflow for temporal and geographic Twitter data. It uses Python and Tweepy to gather Twitter data, and uses Kepler.gl for visualization.


WALLRIDER

Wallrider is a game about falling down walls and shooting ghosts made in Game Maker Studio 2. You can take a look at the source code on GitHub, or download the game for free on itch.io.