Skills and Projects

    Tryndakill was a passion project, and my first ever undertaking of a project outside of introductory coding ones done in school.  In this way, it was my first taste of project building and management.

    The program employed Riot Game's API by pulling, and then parsing, JSONs for live game information.  It would pull character information from the other team, specifically, by giving build recommendation and matchup tips based on the choice of the enemy team's top lane champion.   The program assumed that the player using the program was always using Tryndamere (A champion in the game), and pulled from information provided by a comprehensive guide made by Foggedftw2, a prominent high level player widely considered to be North America's best Tryndamere.  Such information that Fog would suggest in build order and matchup tips were hard-coded in the program, needing only the choice of the enemy top laner to be displayed.

    The GUI for the program was simple, and employed by using Java's Swing framework.   One needed only to enter their username and, if they were in a game, the program would proceed with execution.   JSON conversion was done in Java also.

    The Coronavirus Tracker was my group of three's undertaking for an open-ended final project. It was intricate, and employed many components to create a 2D Map with Pins that displayed and tracked current Coronavirus cases on a state-by-state basis.

  A full review of the project and its intricacies can be read here.

    This project was one of my favorites, and one that featured a number of Machine Learning algorithm implementations attempting to achieve the best classification accuracy of written numbers (pp 1).  I first attempted making classifications based on class means and exemplars (pp 2).  However, I quickly found that a K-Nearest-Neighbors implementation was the one that provided me with the best accuracy, with 98.22% of testing cases being accurately identified with a k of 1.

    Each number can be represented as a 2D plotting in a large dimensionally array regressed down.  Plotting of these for training and then test cases can be seen in (pp 3) and (pp 4) respectively.

    What was interesting was seeing failed cases (pp 5), where incorrectly matched numbers had slight deviations, making them look similar and thus be mislabeled.  Correct matching can be viewed in (pp 6).