Some projects I have worked on. If you are interested, check out my code on Github.
My capstone project at Galvanize, Accent Inspector, identifies foreign accents from audio files of people speaking.
I was initially interested in analyzing music files, as I thought audio would be a cool way to apply the machine learning techniques I had studied in a new environment. After sifting through several music analysis ideas that had already been successfully implemented by others, I came up with the idea to detect accents. This was not something that had been readily marketed and it was a simple enough problem that I felt I could tackle in the two weeks I had alotted to me.
Currently, I am not pursuing further work on this project, however, it was fascinating. Feel free to contact me with any questions, as I have a lot of further insights as a result of my findings.
Gauge was a project for San Francisco's DeveloperWeek Hackathon in 2016. Our team created a public speaking tool for tracking sentiment in real-time.
We used three of the provided APIs (Ricoh 360-Theta, Microsoft Cortana, and HPE Haven On-Demand) for our tool. The app took video during a given speech and uploaded frames to read the sentiment of the facial expressions of the audience. While we tracked audience sentiment in real-time, we also took the audio from the speech, converted it to text, and analyzed the intended sentiment of the speech.
This project was especially relevant as this was an election year! We had a live demo in our presentation and won 2nd place in the hackathon.
This was a case study I completed on fraud detection amongst transaction data. The goal was to accurately identify as much fraud as possible, while maintaining a decent overall accuracy.
This gave me the opportunity to apply predictive modeling to financial data. I had some ideas from my work in the credit card industry about taking note of missing or inconsistent information. After smoting and a lot of cross-validation we were drive recall up to 98.8% while maintaining 76.4% accuracy.
This modeler is a clustering algorithm that organizes NY Times articles into topics using NLP and NMF.
It was interesting to explore the NY Times database through their API. The most common words appearing in the sporting topics were mostly teams, players, and coaches from New York. It seems they prefer to cover their own teams ;)
I built a standard recommender that uses user movie ratings to recommend new movies to a given user. I attempted user-similarity as a means of making recommendations, but the best recommendation method turned out to be a basic popularity recommender on this dataset.
Popples is a command-line game written in Python. Players receive random countries and try to order them by population.
This is a game I had played previously with just a random letter generator to pick country codes, but I figured it could be implemented in Python quite easily. It was a fun project that I debuted at Thanksgiving and played with my family.