Skip to content
learn
Global menu

Guitar Alchemist Lab — Mission

Why this lab

Section titled “Why this lab”

Guitar Alchemist (GA) has a lot of machinery: an index of 313,047 voicings embedded in the OPTIC-K space, chord recognition, fretboard geometry, a 3D neck, generators for images and models. The other courses on this site take each piece apart: the music theory, GA’s AI, three.js, ComfyUI, Blender. This lab puts pieces together into things you can see, hear and hold, and asks each time whether the idea actually works.

Every prototype follows the same contract.

  1. A hypothesis, committed first. Before any measurement, a file in the prototype’s results/ folder says what I expect, with numbers. If I saw a number before writing it down, the file says so.
  2. Measurements. The numbers come from a script in the repository, run on a named machine, with the data pinned: GA’s commit, the file’s SHA-256, the sample’s seed.
  3. Failures published. A prediction that turns out wrong stays in the lesson, next to what happened instead.

Who this lab is for

Section titled “Who this lab is for”

You write C# or Java, you have followed at least one of the GA courses, and you are curious about what GA’s data looks like when you stop reading it through an API. Each prototype says which lessons it builds on. The code is JavaScript for the browser and Node.js, and whatever each prototype needs beyond that.

Prerequisites

Section titled “Prerequisites”
  • Node.js 24 and Git. On Windows, run the commands from Git Bash.
  • A browser with WebGPU: a recent Chrome or Edge. The prototypes fall back to WebGL 2 elsewhere.
  • For the local modes that read GA’s whole index: a clone of GuitarAlchemist/ga and about 200 MB for the index file. The published pages need nothing but a browser.
  • Helpful background: lesson 2 and lesson 3 of the GA AI course, lesson 5 of the IX machine learning course, and lesson 8 of the three.js course.

Outline

Section titled “Outline”
# Prototype Question State
P1 Voicing space explorer Can you see GA’s voicing space in three dimensions, and what does the picture hide? published
P2 Play a chord, see the universe Chord recognition in the browser: FFT, note estimation, template matching over GA’s chord qualities, 3D neck, bracelet and next chord; 624 synthetic strums and 17 CC0 recordings, predictions first published
P3 Album covers Can a diffusion model make a cover that says something true about a chord progression? planned
P4 AI render pass Does a generated texture or lighting pass make GA’s 3D scenes better, measured against the plain render? planned
P5 3D-printable bracelets Can a pitch-class bracelet become a printable object, from GA’s data to a mesh that passes a slicer? published
P6 Full chain Microphone to chord to voicing to image to object, in one run: where does it break? planned
P7 Playability model Can a small model trained on the CPU predict how hard a voicing is to play better than GA’s hand-written cost? published
— Journal

Resources

Section titled “Resources”