Infinite Music AI

Overview

At Infinite Music AI, I worked on MusicGen fine-tuning, audio-text data pipelines and model evaluation. My role covered the full research pipeline, from constructing structured music datasets to fine-tuning models and testing creative interactions with users.

I focused on improving alignment between text prompts and musical structure, and on shifting generative models toward performer-aware and collaborative systems rather than black-box generators. This work sits at the intersection of machine learning, music cognition, and human-computer interaction.

Key Contributions

• Built and curated large music datasets with tempo, mood, genre, instrumentation, and rhythmic features
• Explored contrastive audio-text representations to analyze prompt alignment and failure modes
• Proposed and implemented remixing features conditioned on extracted percussion patterns
• Worked closely with researchers in computational neuroscience and music performance

Evaluation and early product feedback

I evaluated audio-text embeddings across 10,000+ tracks and used embedding and spectral comparisons to guide data selection and model iteration. I contributed to a Discord demo opened to 100+ users, with positive feedback and repeat use from some participants providing early qualitative validation.

Technologies