Ethan Ulrich

• Promoted interest and created education material for ML/AI topics
• Developed leadership and administrative skills

• Educated children from elementary to high school on various STEM topics
• Ran workshops to promote interest and understanding for topics such as DNA extraction, programming, etc

The only participant to tackle both hackathon prompts, earning finalist placements for both ingenuity and best solution awards. I won overall best solution for the advanced challenge by mapping trails from LIDAR data in a mountainous watershed with 99.97% accuracy. For the basic challenge, although the prompt suggested following the path to reach an incapacitated firefighter, I went further—developing a path-agnostic A* search algorithm that dynamically accounted for elevation, using the trail only when it offered the most efficient route.

Outperformed the market by 56% using a daily trading ensemble to predict long/short positions. The ensemble combined XGBoost, RNN, linear regression, DLinear, and a custom CNN/LSTM/Transformer hybrid. Leveraged cointegration analysis, advanced feature selection, synthetic feature creation (including cointegrated signals), customizable loss functions with temporal discounting, and robust evaluation via walk-forward validation and nested cross-validation.

Simulated the encoding and decoding process of Hamming codes and enhanced their error-correcting capabilities by up to 250%. Achieved this by training an LSTM on text data to learn and correct errors beyond the scope of traditional Hamming codes—effectively combining classical error correction with deep learning for improved reliability in noisy communication channels.