Highlights
Pyatt Hall, Vancouver (2014) & ISMIR (2013)
At the Pyatt Hall in Vancouver, a trio performed a program of new compositions by Owen Underhill, Jordan Nobles, James B. Maxwell, and others. Among them was a piece generated by a new AI system I had been working on in my spare time. Professional musicians performed it alongside works by established composers, and it held its place. That same year, I published possibly the first formal framework for quantitatively evaluating generative AI output against human-authored work at ISMIR. The system used hand-crafted feature spaces and a novel statistical method to measure how and why AI output diverged from its training corpus. These techniques, mapping outputs into a shared multi-dimensional space, measuring geometries, and explaining the differences, apply the same comparative logic that modern embedding-based evaluation methods now formalize at scale.
Spliqs: Generative Music, Done Right
The performance from 2014 became the successful proof of concept I needed in order to jump into entrepreneurship. A 10 year partnership with James B. Maxwell was born from a question: could computation augment human creativity rather than replace it? As the replacement path was inevitable. We built a real-time generative music platform focused on the interaction between human and machine, where the AI expanded what a non-musician could do rather than generating content for mass consumption. We were among the first wave of music GenAI companies, alongside Jukedeck, Amper Music, AIVA, and a handful of others, a competitor CEO called it "magical" as how it worked was unexplainable. Spliqs grew well, we got funded and generated 20,000+ hours of music for paying users. Conversations with ByteDance's M&A team after their acquisition of Jukedeck came to a stop as our approach was far from the automatic mass generation style they were looking for. Discussions with major labels to license content stalled, even though we offered attribution, the trust and infrastructure for legitimate licensing simply didn't exist yet. Being too early and too principled became the company's demise.
From Pain to Purpose
Shutting down Spliqs was painful, but it clarified something and cemented an understanding. We had felt firsthand the need for a legitimate path where AI companies and rights holders could work together. Attribution couldn't be an afterthought or a nice-to-have, it had to be at the core of generative AI and not be tied to a particular solution. That lived experience, of trying to do it right and failing because the industry wasn't ready, is exactly what drove us deeper into the problem space, and to start Musical AI.
Winding Path
While generative AI was my focus, I built across a range of domains. In Argentina, I optimized satellite internet servers for the second largest telecom in Latin America, ensuring uptime across the country. At Telus in Canada, I led data reconciliation for the largest migration in Alberta and BC at the time. I developed an NLP-powered search platform that was later acquired. I served as CTO on contract for the final 18 months of a US company, supporting the executive team through to a successful exit. I also enjoyed building mobile augmented reality experiences for one of the largest US weather companies... before mobile AR toolkits had been created.
