00 · Robot Presence Layer

A robot needs to wake up somewhere. SALIX is proof that I can build the layer between technical capability and human feeling: a coherent world, a voice, a spatial score, memory/state language, rituals, and readable intent.

For robotics teams, the relevant point is not "transmedia IP." It is that I built an embodied AI-adjacent world in Unity/Quest with Wwise spatial audio, ElevenLabs voice, LangGraph/pgvector memory logic, and a generative score pipeline from my own catalogue. The same craft applies when a companion robot, home humanoid, autonomous vehicle, or public demo needs to feel legible before it feels useful.

I am not pitching myself as a robotics engineer. I am pitching the missing human-facing layer: authored presence, voice, sonic identity, demo narrative, behaviour legibility, and the world implied around the machine.

World · the machine wakes somewhere

Voice · state · intent

Spatial score · ritual

01 · SALIX

SALIX — a complete transmedia science-fiction universe set in the near future. A 65,000-word novel extended into a Unity VR build, with original music, Wwise spatial audio, ElevenLabs voice, and MidJourney-generated visuals.

A novel, 68-poem collection (Sub-Atomic Blues), with machine learning tools deployed throughout — ElevenLabs voice, MidJourney and Grok visual and video assets, and my own music score implemented in the build with Wwise spatial audio. Toby navigates offworld robotics, consciousness-reading machines, I-Ching decision-making devices, and an environment drawing on contemporary futuristic land art.

The stack is diegetic — its systems are named from inside the fiction (Complete+ save state, PLUS+ memory, HALOW agents) — and the world watches back: room screens caption the player CAM: watching while lines from Sub-Atomic Blues drift through the space as spatial text. Ships as a macOS build and Meta Quest APK.

Complete+ · 01

Complete+ · 02

Complete+ · 03

02 · Spellcaster — Generative Score System

A rights-clean scoring system built from my own finished bounces: 159 source cues were cleaned into 580 prompt/audio pairs, then used as melody conditioning for MusicGen-Melody. The strongest outputs are short cinematic cues with phrase, contour, and texture inherited from the archive.

The system is named Spellcaster — in SALIX, spellcaster booths harvest inner language as product, extraction without consent; here the composer built the booth and climbed in voluntarily. It is an extension of my imagination that directors can play: generate sketches inside my voice, star what you feel — the starred set is the brief, and I write the real score from it. Every cue below carries a provenance receipt. The breakthrough was not another training run; it was finding the right control surface. I compared a base MusicGen model, an earlier V2 LoRA adapter, and a melody-conditioned path using real Simon Taylor bounces as the input. The LoRA generated audio, but failed musically as glitch texture and one-shots; the melody-seed route produced the cues presented here.

Custom Web Audio players · real-time convolution reverb + seamless looping, generated on demand from a deployed MusicGen-Melody endpoint. Batch 2 (twelve seeded cues, June 2026), graded by ear: six selected, one kept as texture, five rejected — every grade logged as training data for the taste layer. Each cue carries a provenance receipt — seed lineage, dataset hash, the artist’s grade.

Does the model carry the composer? Retrieval-with-controls: each generated cue searches all sixteen seed pieces for its own parent — the fifteen non-parents are built-in controls. Two methods were rejected on the way (pitch-class profiles couldn’t discriminate; cross-modal n-grams hit a hub artifact); chroma-DTW survived: six of sixteen cues rank their true parent first (chance: one in sixteen). And the misses are as informative as the hits — Embrace the Serpent ranks 7th on harmony because its inheritance is motivic and spectral, not harmonic: same cell logic, same re-attack habit, development by timbre, a voice-led loop seam.

The hiring signal is practical creative AI judgment: clean source rights, dataset hygiene, model comparison, musical evaluation, and a working path from catalogue bounces to usable generated cues. The failed LoRA path stays in the case study because it shows the system was judged by listening, not by loss curves.

The same training-data philosophy — rights-clean, author-owned, ML-captioned — underwrites the visual-culture forecast at 05 · The Next Five Aesthetic Movements, where the corpus is the CARI / Evan Collins archive and the eventual model output is per-movement SDXL LoRAs.

03 · Burnley FC

Led creative operations alongside the club's first creative director, artist Aitor Throup, for ALK Capital's new ownership of Burnley FC. Office redesign, the Save Football from VAR campaign, full digital shop audit, matchday redesign, and sponsor activations with AG1 and Castore.

Burnley FC · logo · Aitor Throup era · 2024

04 · District Vision × New Balance

Recorded, assembled, and shipped District Vision's first paid audio product — a 16-part course introducing meditation and breathwork as tools for performance gains in distance running. Acquired by New Balance into an ongoing mental-health series.

Inspired by the on-retreat teachings of Esalen Institute founder Mike Murphy, the late Dr. Mike Spino, and George Mumford's The Mindful Athlete. Field-tested in upstate New York and London. Scripted and re-recorded with founder Max Vallot. Published via RSS with cover artwork from Comme des Garçons illustrator Filip Pagowski. Spino's principles — visualisation, breathwork, running on feeling — applied at the Boston Marathon for a personal best of 2:47:17.

District Vision × New Balance · Center for Inner Peace

05 · The Next Five Aesthetic Movements

A two-horizon taxonomy of where visual culture is moving next. Volume I (March 2026) names five movements for 2026–2031, derived from the Evan Collins / CARI corpus of 5,395 catalogued artefacts across 38 aesthetic categories. Volume II tests those propositions against 1,266 institutional exhibitions surveyed at seventy venues in London, New York, and Berlin, then extends the forecast forward to 2042.

The framework operates as a morphological forecast: not speculation, but structural inference. Five complementary lenses converge on five aesthetics that are not yet here but are, in the strict sense, already structurally inevitable. Every prediction is paired with a 2028 falsifiability test.

"The exhibition is undergoing a phase transition from visual consumption to somatic negotiation — from cinema to pharmacy."Archaeology of Attention · Part One

06 · Music & Film Scoring

20 years composing — half a billion streams as a songwriter and producer. Commercial composition for Nike, Airbnb, Meta, and New Balance via leading music houses, including Airbnb Plus' global launch films in Paris, Mexico City, and Bath. Collaborated with Rihanna, Mark Ronson, LCD Soundsystem.

From commercial composition for global brand films to spatial audio, galleries, and immersive installations — this is the taste base behind the technology work.

Commercial Composition Engagements

Nike · Airbnb · Meta · New Balance · District Vision · Audi · Boss

Tools

Ableton Live · Pro Tools · Logic Pro · Spatial Audio · Field Recording · Modular Synthesis · Film Scoring · Vimeo