What we have built.
Products, public tools, personal projects and client work, almost all of it since Sisyphe began in 2023. Each card is tagged with the layers it works hardest on: perceive, reason, act. For the two decades before, see the background.
Music production
Tools that work inside your DAW, on your session, with real undo.
Sideman
Gives Claude the whole of Ableton Live through Live’s own object model. Observers report what you do by hand, a skill teaches the model how Live’s paths are shaped, and thirty tools write into the Set with real undo. About Sideman →
Roadie
From finished track to Live set. A recording of your own track or set in, a Live session out: bass, drums, synths and vocals separated, drums split six ways, and the stems loaded into your running Set through Sideman, on the bar line, ordered by frequency, colour-coded and levelled, with a locator per section. Every default measured on real material. Roadie site →
Agents & simulated worlds
Embodied agents in simulated worlds, perceiving what is around them and acting on it: an ecosystem that composes, a simulator that flies.
Dream Fauna
A real-time generative-art instrument. Watercolour creatures, lifted into 3D by an AI pipeline, live in an artificial ecosystem whose swarm composes the music. Food webs and boom-and-bust cycles emerge from who can perceive what, and in the live set the music drives the swarm back. About Dream Fauna →
Bestiary
A MuJoCo simulator for things that move. Its first creature is a quadrotor that speaks MAVLink, so flight scripts written for the Udacity course simulator connect unchanged, headless at 30 to 60 times real time. Alongside it, a sim-to-real project on NVIDIA Cosmos.
Creative ML
Generative models that set paintings in motion, and the audio toolkit underneath the music work.
Moving Paintings
Still paintings dreamed into motion with diffusion models. Each piece takes one artwork as a seed and lets its brushwork, palette and forms flow, morph and recompose: seven pieces, from Bosch to Basquiat. Watch the series →
Nimrod
A personal open-source toolkit for speech, audio and language modelling: modules, models and training recipes for getting deep-learning experiments running quickly. Nimrod docs →
Learning & culture
Tools that help people learn by doing, and find what is on in their city.
Pappus
A local-first notebook that teaches. The AI sees every cell above your question and answers a hint or a guiding question at a time. It never runs a cell, and by default it cannot write files: running the code stays the learner’s job. Built on Answer.AI’s SolveIt method. Pappus site →
Westside LA Events
Sixty scrapers over venues, museums and community calendars on LA’s Westside, unified into one search with maps and analytics, deployed on Cloud Run. A city as an archive: the perceive layer, at civic scale. Open the app → It may take a few seconds to wake.
Client work
Engagements for companies building AI into creative, cultural and voice products.
Tamada × Visto
An AI curation engine for Visto, the art consultancy that builds collections for luxury hotels, residences and landmark destinations: search a collection by text or by image, on CLIP embeddings and a vector index, taken from prototype to product with a team.
Persona
Real-time conversational voice pipelines, built inside the product team.
Parts Order
Scientific advice on music source separation.
Start a conversation
Tell us what you are building and where it is stuck. You will get a reply from the person who would do the work.
Or copy the address: sylvain.legroux@gmail.com