When people design any product that involves running fluids through a porous surface (like air filters, athletic wear, or a sprinkler), they like to keep track of a number of factors that will dictate how that fluid will run (like the surface's surface area, roughness, waviness). One particular factor they like to measure is 'tortuosity', a measure for how 'complex' a fluid's path will be from one end of the surface to the other. My goal was to use one of the university's datasets ( a collection of labeled images ) to help predict that tortuosity. I mainly used a convolutional neural network (using PyTorch) for this, looking at the images pixel-by-pixel. But I also used lighter models, too, like linear regression models and decision trees. All in all, a super fun introduction to the field!

I used a RNN to create a speech recognition model and trained it on a thousand hours of the CommonVoice English dataset. I also showcased it in a Django web app, so you can see how the model 'thinks' about its input! Here is the link to the demo:

I built a transformer model with 12 layers, each layer having 12 heads in the self-attention stage. I downloaded some text from the CommonCrawl dataset and spend $25 training it in the cloud. I then incorporated it into a Django web app, easily letting you see the response and the predicted tokens at each stage. Here's a link to the demo:

I worked in a 12-person team following the Agile methodology. We constructed a web application in React, MongoDB, and Tailwind that would collect data from the users and run it through a set of analytics to display to the company's administrators. This was a cool introduction to the field of software development! My tasks were usually in the space of working on CRUD operaitons and HTTP requests to our database.