Video search
Video Moment Finder
Semantic and transcript-aware video search for finding exact moments inside long videos. Search with text or images, then jump straight to the relevant timestamp.
View repoJuan Carlos
I’m Juan Carlos, a Senior ML Engineer at Marvik. I build applied machine learning and generative AI systems, with a focus on moving research ideas into products people can use.
I studied music before working in AI, and I still think a lot about creative tools: what happens when machines can compose, generate, search, and collaborate with us.
The thread through my work is ML engineering at the intersection of music, creative AI, and production systems.
Selected Projects
Music AI
Multitrack MIDI Music Generator
A Hugging Face Space for generating multitrack MIDI by genre, built from symbolic-music experiments with custom datasets, tokenization, and model fine-tuning.
Open demoResearch notes
Diffusion from First Principles
A 52-week research curriculum rebuilding diffusion models from probability foundations through score-based models, SDEs, and flow matching.
Writing
Research Log
Technical Notes
Short, question-first notes on papers, derivations, experiments, and ideas from personal machine learning work.
Browse writingTutorial
Using Hugging Face to Train a GPT-2 Model for Music Generation
A walkthrough for building a symbolic-music generation workflow with a custom dataset, tokenizer, GPT-2 training loop, and interactive demo.
Read article