John V Flickinger

Chess.com

By far one of the most impressive projects I have built to date is still up on chess.com here. Recently graduated and while freelancing I was hired to integrate my other project hellochess (also in this portfolio) into chess.com as a proof-of-concept. ofcourse the initial version was not nearly as refined as the current version but even then most of the core features were there including drawing arrows during Team games.

Close Window

Hellochess & Tetra AI

Hellochess is a website I built after having a year long cooperative education hiatus from college. While at the job, one of my coworkers brought a 4 player chess board to work and everyone in the office took turns playing. When looking for a way to play online, I found that there was no way to play 4 player chess online. So I built Hellochess on my spare time. The project evolved several times from PHP to Node.js and finally to React.js. The website eventually had a small community of players and was a great way to learn modern web development. Several enhancements were made to the website including a 4 player chess engine that I built from scratch. The engine is called Tetra AI and is a minimax algorithm with alpha-beta pruning. I had some interest from the chess community and a fellow developer named Joe, also added schess to the website. Here is Joe's github if you want to check out his work.

Close Window

Overcooked AI

In Fall 2023, I took a Reinforcment Learning course at Georgia Tech, which has been my favorite course in OMSA so far. The course was created by Dr. Charles Isbell and Dr. Michael Littman and was quite challenging. There were many great projects in the course, but by far the hardest was the final project. For fall 2023, the final project was unique as it was the first time the instructors introduced a new multi-agent environment. The environment was based on the game of Overcooked. The goal of the project was to create a pair of AI agents that could work together to solve multiple layouts. The game was quite challenging as the agents had to learn to work together. Ultimately, I was only able to solve 2 out of 5 layouts but I was able to get the agents to work together. The agents were trained using the A2C algorithm, but I got the chance to experiment with other algorithms such as DQN and SARSA. The project was a great learning experience and I hope to continue my exploration of reinforcement learning in the future.

Close Window

Controller

From early 2016 till my graduation in May 2017 I worked for Dr. William Yeoh on a National Science Foundation funded project. The project, which we named MESH, (Mult-agent Energy-efficient Smart Homes) was a distributed Smart Home project with the aim of reducing peak power consumption usage (e.g. the time of day when energy costs the most). The project was awesome, unfortunately, the team met several hickups and one of them required a re-write of our entire code-base. After I realized most of my code would need to be re-written I decided to re-write my PHP/Codeigniter application into NodeJS and Express. Long story short my project was extremely dependent upon changes to each house agent controlled in Java and constant changes to the database schema resulted in medial progress. Ultimately a basic version of the application was finished and I am really happy with the resulting application that used TypeScript, React, Redux, and NodeJS. If you want to read more about the project please visit this website (which I also built).

Close Window

Experiments with Mancala

I have a long history with the game of Mancala. I first learned the game on an old Nokia mobile phone and have been fascinated ever since. During my undergraduate AI course our teacher challenged the class to a mancala tournament using 6 pebbles per pit. I was able to win the tournament convincingly using GMKalah, a mancala AI I built in C using Alpha-Beta pruning and a strong evaluation function. I have since built several Mancala AI's using different algorithms and languages. My primary goal has been to build a muzero Mancala AI but with minimal help from the internet. In 2017 I was able to leverage opensource code to build an AlphaZero Mancala AI, but it was unable to beat GMKalah and also leverage much code that I did not write. I have since built a more original AlphaZero Mancala AI using Python and Pytorch with additional features that graph the Monte Carlo Tree Search as the AI expands the search space, but ultimately my goal is to enhance the algorithm with MuZero and finally dethrone my old GMKalah AI. All this ofcourse, is done as a hobby, but the project has been a great way to learn about reinforcement learning and planning algorithms.

Close Window

Aura Health Freelance Project

As a way to demonstrate my css capabilities, here is a freelance project I worked on for Aura Health a mediatation app. The website supplemented the app by allowing teachers to upload meditations and choosing emotions, audio files, course title, etc... This project is special in that it uses Next.js for the front-end and firebase for the backend.

Close Window