Tommy Bird

Hello, I'm Tommy Bird

And this is my Computer Science portfolio.

About Me

I'm a Computer Science student from San Francisco

I am currently studying at Rensselaer Polytechnic Institute, pursuing a Bachelor's in Computer Science.

My strongest technical skills are in Python and C, but I am also well versed in Java, Swift, JavaScript, SQL, and many other languages and technologies listed on my resume. I have a strong history of leadership and communication skills and tend to get involved wherever I can – such as my role as a project lead for my school's open source club.

Below are some projects I've worked on.

Projects

Portfolio

Here are my projects:

I have worked on a variety of personal projects, from simple games to complex mobile applications. Below are the ones I believe are worth showing off. If you want more information on any of them, either look at my resume or visit my github!

On the left are the two input images. On the right they are fused together at matching points.

Image Blender

Python Computer Vision Program

If you have multiple images of the same scene taken at different angles and/or positions, this program will blend any number of them together, creating a higher resolution image. This was achieved with computer vision techniques such as SIFT, Flann-based matching, RANSAC, and a good bit of matrix algebra on homography/fundamental matrices. It was made in Python, leveraging the OpenCV and NumPy libraries.

Source Code

After selecting the term "Arctic Fox," the extension displays an AI-generated contextual definition (left), as well as an interactable 3D model of the fox (right).

Readee

Chrome Extension

Readee addresses a fundamental challenge in online learning and reading comprehension: the difficulty of visualizing and understanding complex terms and concepts. The tool was created for a hackathon within a 24-hour constraint. It uses Echo3D's JS SDK, and GPT-4 API via Node.js hosted on an Amazon AWS EC2 instance. The front-end and button injection was made by using HTML, CSS, and JavaScript.

Source Code Demo Video

Here is the output when placing 300 people via blue noise into the floorplan on the left. The areas marked green are the most visible wall-spaces (where you should place art), and the red are the least visible.

Art Optimizer

C++ Program

This program takes in an input of a floorplan, and outputs a heatmap specifying where it is best to place art based on visibility. It uses the CGAL library to generate a Minkowski sum of the floorplan, then triangulates it using a constrained Delaunay triangulation algorithm. It then uses raycasting to determine point visibility, with options for both white noise and blue noise placement.

Source Code Research Paper

These are simply (outdated) screenshots of the gameplay, showcasing a birds-eye view of the randomly generated map, as well as the throwing mechanic on the disc.

Daily Disc

Mobile Game

Daily Disc is a mobile game I created in the Godot game engine. It boasts randomly generated maps, created using perlin noise. The game features a leaderboard system as well. Each day, a daily course is given to all users, and players can compete to do better than their friends.

Demo Unavailable

Here you can see the three ML agents (a wizard, priest, and warrior) fighting the boss character.

RLOTMG

Cooperative Multi-Agent Reinforcement Learning System

RLOTMG is a demo that features cooperative multi-agent reinforcement learning in a 2D bullet hell game inspired by Realm of the Mad God. It was made using Unity and C# as well as PyTorch and Tensorflow via Unity's ML-Agents toolkit for training. There is also a research paper that goes into more detail about the project; also a demo can be found on repository's github page linked below.

Research Paper Demonstration

This image shows an example conversation with the tool, wherein the user is asking questions about the input video. The quality score is displayed in the left, denoting how well received and not click-bait the video is out of 10.

Vieworthy

AI Video Curator

Vieworthy is an AI-powered video curation website and Chrome extension that analyzes video data evaluate quality and provide users with video summaries. It is capable of web scraping for Youtube videos, and processed said data to query an LLM API where a description is generated. It was made using Node.js, OpenAI's GPT-3.5 API, and Google's Youtube Data API.

Source Code Visit Website

This is very outdated image showing early developmental screenshots of the social media, route generation, and profile interfaces of the app.

Path Pilot

iOS Application

Path Pilot is a fitness-based iOS Application that displays randomly generated routes for runners. Made with SwiftUI, it uses the Mapbox API with an updating database hosted on Firebase. This project is still in development.

Show Demo

This is the homepage, not much to it other than it shows green-spaces in Troy. Useful for finding a good picnic spot.

Troy Green Spaces

Website

Troy Green Spaces is a website that displays an interactable map of Troy, NY with a list of all its green spaces. This project was done with an environmental science student at a nearby school based on her research. It was made using HTML, CSS, JavaScript, and Leaflet.js. GIS data is fetched from the City of Troy's open data portal.

Visit Page

This screenshot shows the level map (green is valid tiles, blue is water), as well as all the waypoint nodes (dots). After selecting two dots, an optimal path will be found between them, marked in yellow (dots marked red were those explored in the process).

A* Pathfinding in Dragon Age

Unity Demo

This is a Unity demo that uses the A* Search and Bressenham's line algorithms for pathfinding on maps from the series "Dragon Age." It allows you to choose a map, a heuristic, a waypoint representation, and any two points.

Run Demo

This is a gameplay screenshot, you type the answer to the question and your score (as well as the question difficulty) increases. It's space themed because that was a bonus category in the Hackathon. Unfortunately we didn't win anything because the version we turned in rendered like 20 planet gifs all animated at the same time and it'd totally fry your computer - whoops.

Astromath

Webgame

This is a simple webgame myself and some peers made using HTML, CSS, and JavaScript. It was made for a Hackathon, and was created within a 24-hr constraint.

Source Code

You're quite literally viewing the code for this particular section in the page right here.

This Website

Website

This website was made using HTML, CSS, and JavaScript. It is hosted on GitHub Pages, and the source code can be found below.

Source Code