Companion was designed to help musicians practice and perform with the feeling of a full ensemble, even when they were alone. The system listened to a soloist in real time, followed the score, and adjusted its accompaniment naturally. As a pianist for over ten years, this idea immediately drew me in. I had spent years practicing with recordings that never responded to me, so the chance to build software that did that felt both practical and exciting on a very personal level.

For the project, I helped design evaluation frameworks, test synchronization accuracy, and develop methods for handling tempo shifts, rubato, and unpredictable timing. I also researched dynamic time-warping, PID control, and multi-performer interactions, and constantly tested the system on real musical pieces.

The lab environment was highly collaborative. We shared scores, recordings, and experimental results, and iterated quickly based on what we heard and measured. The work blended machine learning, music theory, and audio engineering, and it was rewarding to see musicians react to a system that felt more like a partner than a tool.

A key highlight were the conferences. I served as lead presenter twice, for Purdue's Undergraduate Fall and Spring symposiums in 2024-2025 where I spoke to both researchers from the region and the general public.