Tansen / Copilot
AI/ML engineer
Remote · Full-time · 1+ years
Build models that edit structured music projects, working with tracks, sections, and MIDI so a change still respects what the producer asked to keep.
Melodex keeps the song as a project. Tansen is the model layer that has to understand tracks, sections, MIDI, and a request like making the chorus bigger while leaving the vocal melody alone.
This role is for someone who wants to make models that edit music. You will work on training, evaluation, and inference, and on the path that turns a producer’s request into a scoped, undoable change inside the studio.
The interesting problems are musical. Keeping a melody intact, changing the energy of one section, and checking whether an edit still holds together two minutes later matter more to us than a prettier demo clip.
What you will do
You will train, evaluate, and ship models that work on structured music, from notes, rhythm, and harmony through arrangement and section-level edits. Copilot should turn a natural-language request into a musical plan that respects what the producer asked to keep.
Evaluation has to go further than whether something sounds fine. You will look at whether constraints were followed, whether the song still holds together, and whether the edit is actually usable. The data is real session history, including accepts, undos, tweaks, and the gap between a good loop and a finished arrangement.
You will work with the desktop team so the models land in the product as tools producers can use.
Requirements
We are looking for at least a year of machine learning engineering or research, with hands-on Python and a modern stack such as PyTorch. You should have trained or fine-tuned models yourself, and you should be able to set up evaluation, look at failures, and keep iterating. The team is small and fully remote, so you will own a problem from start to finish.
Good to have
Work in music information retrieval, symbolic MIDI, audio models, or sequence models is useful. So is experience with transformers, diffusion, or other generative models, and with production inference, latency, and shipping a model into a product. Music theory, composition, time in a DAW, or generation that has to follow constraints will help.
What we offer
Pay is competitive, and it follows the work. We also cover a home setup, so you can put together a proper desk, monitor, chair, and the basics of a working studio.
The roles are completely remote. There is no office to relocate to and no days-in-building requirement. The team is small, the product is a desktop studio producers already use, and you will own problems from one end to the other.
How to apply
There is no application portal. DM @MelodexStudio on X, say which role you want, and include a short note on why the work interests you. Add a link to something you have built, whether that is GitHub, a portfolio, a paper, a track, or a demo. A concise message is enough. We read every one.
Apply through X.
DM @MelodexStudio with the role title, a short note, and a link to your work.
Open @MelodexStudio