Loan Default Risk Analysis

Loan defaults are not only bad for business but are also stressful for the loan taker. Defaulting on a loan not only means the bank lost money, but the client’s credit score has dropped significantly, and it is unlikely a bank will be willing to lend money to the person in the future. If it is possible to predict and maximize the repayment rate of loans, this would be a great benefit to both parties in the exchange.

The data for the this project was extracted from Kaggle's Home Credit Default Risk Competition. My team applied a variety of advanced algorithms to engineer a machine learning model for predicting loan default risk. Some of the advanced methods used are: Feature Engineering, LightGBM, and using GridSearch for optimization.

We were able to achieve 75% prediction accuracy in the Kaggle competition, while the first place winner scored 80%

Google SPS

Google’s Software Product Sprint (SPS) is an 11-week invite-only program for high-potential students. During SPS, my team developed a web application for finding compatible roommates using Java, JavaScript, and leveraging Cloud Datastore to store user profiles. We collaborated with Software Engineers at Google to use best practices and overcome challenges.

During my Google internship, I served as a panelist for the next iteration of the program to share my journey.

BattleSnake

BattleSnake is a national hackathon where developers across Canada form teams to build AI snakes. At every game, a number of snakes compete on the same board, with the winner being the last one standing.

My team's AI was able of defeating multiple competing snakes, as well as a variety of bounty snakes, which are snakes created by industry sponsors.

The snake has a variety of game strategies, playing defensively and chasing its tail when the board is crowded, and taking risks when it only a few snakes remain. It also uses graph coloring algorithms to predict if it will get stuck.

Engineering Robotics Competition

The Engineering Robotics Competiton is an annual, semester-long competition held for engineering students where the task is given at the start of the semester. For 2019, the task was to create an autonomous robot to tether cables to IR beacons, serving as a prototype for an underwater rover for internet cable tethering.

Our robot was the fastest to complete all of the competition tasks. The robot had various optimizations, such as dynamic speed settings based on the distance from the target, as well as adjusting motors unevenly to steer the robot. The robot was programmed using RobotC (a variant of C).