Our first Galvanize Data Science hiring day is coming up at our Denver – Platte campus, and we’re excited to see students unveil what they’ve been working on for the past few weeks. If you’re not familiar with Hiring Day at Galvanize, think of it as a demo day for students where they show off technologies they’re learned, projects they’ve built, and get a chance to meet with potential hiring partners.

Hiring Day is free to attend, so if you’re looking to add an awesome data scientist to your team, register now

Here’s a quick rundown from students about their projects:

Germanwings Air Crash Media Analysis: Hugh Brown The Bit.ly data of global response to the Germanwings air crash in March 2015 provides invaluable insight into internet activity. Through the magic of data visualizations and topic modeling, I explore how global interest varied by country and timezone as the story emerged and changed.

Using GPS to Create Workout Leaderboards: Cary Goltermann Strava tracks users' workouts via GPS to compare athletic performance in running, cycling and other pace-based workouts. The metrics that Strava collects allow them to keep an updated leaderboard for users' performance on a segment-wise basis, but they currently have no notion of an overall leaderboard. My project uses the magic of linear algebra to build a leaderboard that creates a leaderboard even for riders who have ridden in different places.

Analyzing Resource Demand with Device Sensors: Chris Seal I am building a system that uses data from sensors to determine what devices or appliances are currently turned on and the resource demands attributed to each device.

Predicting Customer Purchases with Machine Learning: Deseanae Bluiett My project is "More than Guts!" and uses Data Science to easily provide data driven sales and marketing insights. The focal point is an interactive sales and marketing dashboard that identifies customer segments and utilizes predictive model to predict the likelihood of product purchase. I utilize machine learning techniques to create and optimize the predictive models in addition to unsupervised learning techniques to create the customer segments. To create the app interface, I utilize Flask, Pyxley, Bootstrap, HTML, JAVA, CSS and Python. Though data segmentation and predictive modeling is traditionally applied in business cases, there are plethora of cases where these methods could be beneficial. To illustrate, I created additional predictive model which alerts users when they are entering a potentially dangerous zone based on crime data, gps location and time of day.

Recommendation System for Packages Using Haskell Programming Language: Matthew Herzl Using the correct set of component packages is an essential part of programming. However, for any given functionality there often exist multiple package options, and exploring the usability of each contender on its own to make the most educated decision on which to use can cost valuable coding time. A programmer might use Google or word-of-mouth and try one package after another until s/he comes across one which provides the desired functionality. This method works to an extent, but leaves room for more specific and accurate solutions which narrow search constraints to packages within a particular desired language and have the full set of potential options at their disposal, and assist a programmer to consider the various options and efficiently determine which to use. My project, which is a recommendation system for packages written for the Haskell programming language, approaches this package-search problem directly. It uses the full set of Haskell packages (Hackage) and data science techniques to estimate the quality and functionality of each package, thus addressing the package search problem directly.

Media Coverage Analysis During the 2016 Election Cycle: Thomas Brawner The project compares coverage of the 2016 U.S. election cycle across three major international news outlets, The Guardian (U.K.), The New York Times, and The Wall Street Journal. The first component of the project is a classification model in which the objective is to predict the source given the article content. The second component uses topic modeling to shed light on how the variation in content across the three sources permits accurate classification.

Using Neural Networks to Read and Understand Arbitrary Text: Sami Zemedkun I am using long short-term memory (LSTM) recurrent neural networks (RNNs) to create a system capable of reading arbitrary text and answering questions about that text. I am testing my system on Facebook’s bAbI Q&A dataset. The model at this moment is more accurate on the Facebook bAbI dataset than any other results I am aware of.

Identifying Forest Fires in Real Time with Satellite Imagery: Sean Sall Currently, NASA satellite imagery of detected fires is not used in real-time for forest fire prevention. One potential reason is that the detected fires from this data set contain a high number of false positives, ranging from hot asphalt parking lots to house fires to farmer burn piles. I aim to build a model that can identify which of these detected fires are forest fires, using historical fire perimeter boundaries and weather data. My hope is that I can accurately identify these forest fires so that the detected fires data set can be used in near real-time to aid forest fire prevention, where every minute counts.

Predicting Bitcoin Price Fluctuations: Chris Bynum My project aims to make high frequency bitcoin price predictions from market microstructure data. The dataset is a series of one-second snapshots of open buy and sell orders, combined with a record of executed transactions. The data is being collected live and currently consists of 2+ million observations. I aim to predict price movements over 30-second intervals and deploy an app displaying live predictions and model performance. Register to Attend Colorado’s First-Ever Data Science Hiring Day