DERIVE

2024 · code, space, systems

Memory can be retrieved through felt proximity without forcing fragments into a clean chronological story.

Source fragments are transcribed, divided into traceable units, embedded, and arranged in a vector field where similarity creates possible paths. Human-authored tuning rules resist automatic smoothing and preserve rupture, uncertainty, and provenance. Retrieved material can remain navigable or drive an external sonic environment through OSC or MIDI. The project does not claim to model trauma clinically; it explores how computational retrieval can preserve discontinuity without converting memory into false coherence.

Cinematic Rail Type: 12-card lead system rail
Pipeline: source, transcription, chunking, embedding, retrieval, and performance signal
Constraint: provenance and discontinuity remain visible throughout generation
Status: audio and vector-processing experiments documented; public demonstration evolving