Portfolio

Natural Language Processing

Building and training a custom RNN

This Mad-Libs was completed by a recurrent neural network trained on 183,146 court cases from the state of Illinois downloaded via the Caselaw Access Project.

I created this document in order to showcase how much my custom neural network had learned. Using three 128-cell layers of gated recurrent units, I was able to train the neural network to predict the next character in a sequence succesfully enough that it (mostly) learned how to spell! I'm sure that one day Black Mirror will write a script in which their robot also wants to "go hunting for the plaintiff."

Click on this link to download my powerpoint presentation.

• Date: Spring 2020
• Class: Machine Learning

Close Project

Tic-tac-toe Robot

Engineering & CS design project

This video shows the proof-of-concept prototype that I constructed using a lot of hot glue, duct tape and zip ties. Building this gave me many ideas for potential design improvements, too bad I don't have more time on this project!

I wanted a challenging yet achievable project that required mechanical and electrical engineering as well as sophisticated machine learning techniques. This project was perfect because I got to write image processing software and controls software, all while designing and building a mechanical arm. I learned a ton about the engineering design process, and I hope that it can inform and inspire future students!

Completing this project was my biggest challenge so far, as it turned out to involve much more work than I expected. My final design report includes full design drawings with CAD models and wiring diagrams, as well as detailed specifications and a request for proposals with standard contract. Since I am the only student ever to graduate Carroll with an Engineering and CS double major, I didn't have any team members which only added to the challenge. Ultimately, this was the perfect capstone to four years of my life in which I learned and grew a truly unmeasureable amount! I can't thank my advisors and professors enough for all of their help and support.

Click on this link to download my final design report, and this link to view my recorded presentation.

• Date: Spring 2021
• Class: Senior Design

Close Project

Climate Change Visualization

A big data project using NWS storm records

This animation shows storms in the U.S. over the past 60 years, colored by storm type and scaled by the number of deaths caused, based on records from the National Weather Service and the National Oceanic and Atmospheric Administration.

I created this graphic in order to display one aspect of climate change in a visceral, tangible way. Given that there are many potential causes for the explosion of storm-related deaths in the U.S. in recent decades (population growth, record-keeping improvements, etc.), I leave the exact role of climate change to the viewer's discretion. However, I hope that this animation can help instill in viewers the magnitude of the threat we face, and at the end of the day, inspire positive change.

Click on this link to view my full report.

• Date: Fall 2019
• Class: Adv. Stats. for Data Science

Close Project

AM Radio

Constructing a single-sideband modulator

This oscilloscope capture shows our single-sideband output function (red), formed by modulating the amplitude of our carrier wave (blue).

We used a diode bridge limiter and Op-Amp to set the amplitude of our carrier wave based on input from our controlling function, a translated sinusoid. We then used a resistor and capacitor in parallel to recover the original controlling function from our amplitude modulated carrier wave, demonstrating the capability to send and receive information via an AM radio wave. During our post-lab discussion, we analyzed a similar design for a double-sideband modulator using two diode bridges.

Click on this link to download my full report.

• Date: Spring 2020
• Class: Circuit Design & Analysis II

Close Project

Black Box

A phasor domain circuit analysis challenge

This plot shows our gathered data compared with our upper and lower estimates for the contents of the black box.

We were tasked with identifying the contents of a "black box" containing two passive circuit elements, connected either in series or in parallel. By analyzing all of the possible combinations, we were able to find their complex impedances as a function of the driving frequency ω. We then constructed a circuit and measured both the voltage across, and current through, our black box at different frequencies. By converting the voltage and current into complex phasors, we were able to find the equivalent impedance of our black box at different frequencies. Using our gathered data, we estimated that our black box contained a resistor and capacitor in series of 10.5-11.1 kΩ and 3.02-3.19 nF, respectively. While we correctly identified the combination of elements, our resistor and capacitor were actually measured at 11.3 kΩ and 2.19 nF.

Click on this link to download my full report.

• Date: Fall 2019
• Class: Circuit Design & Analysis I

Close Project

Gear Rental Database

Designing a relational database for inventories

This entity relationship diagram illustrates my design for a database to track outdoor gear rentals and returns.

I designed this database for the Carroll Adventure and Mountaineering Program, where I work. The idea of the project was to use a database to track outdoor gear rentals in order to be able to view current available gear and current rented gear, as well as automatically email return reminders and assess late fees. The database would also keep track of rental histories, and would therefore enable statistical reviews to better inform choices such as operational hours, quantities of different gear, and when to replace items.

Click on this link to download my powerpoint presentation.

• Date: Fall 2018
• Class: Database Design

Close Project

Gravity Wave Detection

Atmospheric Research in Villarrica, Chile

Animation of a wavelet transformation applied to time series data.

Atmospheric gravity waves are an important mechanism for the propogation of energy throughout Earth's atmosphere. However, for a phenomenon that affects broad-scale weather and climate patterns, the mechanisms of generation are not well understood. For example, the thermal disequilibrium of a total solar eclipse is predicted to cause waves, yet this has only been experimentally verified once, in 2017.

In order to corroborate these findings, I worked with a research group from four states funded by NASA and the NSF to travel to Chile during the midst of the COVID-19 pandemic and collect 48 hours of continuous atmospheric data across two sites, the longest comprehensive radiosonde data set ever recorded. To analyze our data, I developed software using the wavelet method to detect waves in the vertical radiosonde profiles collected by our weather balloons.

Click on this link to read a NASA press release, this link for a recorded presentation on the wavelet method, and this link to view the project on github.

• Date: 2020 - 2021
• MSGC, ISGC, NASA, NSF

Close Project

Exposure Time Calculator

Software for W. M. Keck Observatory

An exposure time calculator for multiple instruments at Keck.

Exposure time calculators are an important part of astronomy preparation. These planning tools are used to estimate an exposure time from the desired resulting signal to noise ratio, or vice versa. This estimate depends on many factors, from instrument-specific parameters such as filter or grating to atmospheric parameters like water vapor.

I got the honor of designing and building an ETC, including a GUI and API, for many of the instruments at Keck. In addition to full stack web development, this included learning about astronomy and the physics behind it, and designing a programming structure that could be easily modified to add new instruments. This was a super interesting project where I learned a lot that was such a joy to work on!

Click on this link to see the project documentation on github.

• Date: 2021 - 2022
• W. M. Keck Observatory

Close Project

Mathematical Contest in Modeling

Intense 100 hour international competition

A simulation of two-dimensional hyphal growth for competing species of fungi in various environments.

The Mathematical Contest in Modeling is an annual competition that draws undergraduate students from around the globe to solve open-ended questions in a single intense, 100 hour session. Working in teams of three, students must pick a prompt from among the six options available, then research and create a mathematical model to answer the question. At the end of the weekend, the final deliverable is a 20 page paper that justifies all assumptions made and evaluates the model with an in-depth sensitivity analysis and suggestions for real-world application.

I competed in the MCM in 2020, where my team modeled the impact of climate change on the scottish fishing industry due to fish migration with changing temperatures. In 2021, we modeled the growth of competing species of fungi in various environments, using our model to examine how biodiversity affects decomposition rates. This year, our submission recieved an honorable mention, placing us in the top 33 % of the 10,000 teams that competed.

Click on this link to download our 2021 final submission.

• Date: Spring 2021
• Carroll College

Close Project