My studio is built on two things: creative direction and structured systems. Every project sits at the intersection. Video production, software engineering, AI education. Same brain, two hemispheres.
Invisible connections between domains is where the real value lives. That principle runs through everything I'll show you today.
The Result
One month. Zero acoustics background. 14 differential equations.
A physics-accurate singing bowl synthesizer running on an iPhone. Real-time Cymatics visualization with Metal shaders at 60fps. Physarum simulation with 500,000 particles.
Live demo: Touch Grass app on iPhone + Bluetooth speaker.
But first, AI lied to me.
What AI gave me
Looks plausible. Completely wrong.
Invented power law. Real bowls don't follow n^1.8.
What the book says
Inacio (2006): empirical ratios from laser vibrometry
No universal formula. Each geometry has measured ratios.
The model was confident. The physics was wrong. Without the book, I would have shipped garbage.
The Method
How did I get here? Four steps.
Step 0
Find the books
Deep Research: describe the problem, get canonical references.
Step 1
Extract the knowledge
MinerU: OCR + LaTeX from scanned PDFs.
Step 2
Index and search
RAG: 5,000+ chunks, search by meaning.
Step 3
Connect across domains
Obsidian: wikilinks, invisible connections.
Step 0: I didn't know which books I needed.
My prompt
"I'm building a physics-based synthesizer for singing bowls. I need the actual mathematical models, not approximations. Which academic papers and textbooks contain the measured modal frequencies, damping models, and radiation coefficients?"
10 min later: Inacio (2006), Bilbao (2009), Fletcher & Rossing (1998), Rossing (2000).
Step 1: Extract the knowledge.
Most academic books are scanned PDFs. Formulas are images, not text. ChatGPT file upload loses the math. MinerU preserves LaTeX.
"How does damping vary with frequency?" → returns the exact Bilbao formula, page 247.
Step 3: Connect across domains.
Acoustics→Modal Frequencies→Bessel Functions
Psychology→Entrainment→Brainwave States
Hearing Loss→Frequency Response→Acoustics
My son has hearing loss. Researching his condition led to frequency response curves, then acoustics, then this app. Obsidian made these connections visible.
The Application
Once you have the knowledge, what can you build?
Beyond writing code, AI agents can produce deliverables: presentations, visualizations, reports. Let me show you two approaches.
"Make me a presentation."
Apple Keynote has a scripting API. The AI agent creates slides, sets titles, adds content. You describe what you need. It appears.
// What I said: "Take the masterclass content and put it into Keynote. Minimalist Dark theme." // What the agent did: tell application "Keynote" make new document with properties {document theme: "Minimalist Dark"} make new slide with properties {base slide: master slide "Statement"} ... end tell // Result: 22 slides in 30 seconds.
Live demo: I'll describe a slide, and you'll watch it appear in Keynote.
"Make me a video with animated formulas."
Manim, the engine behind 3Blue1Brown. LaTeX formulas animate on screen, graphs draw themselves, equations transform. The agent writes the Python scene, renders the video.
# The agent writes this: class BowlEquation(Scene): def construct(self): formula = MathTex( r"y(t) = \sum A_n e^{-\sigma_n t}" r"\sin(2\pi f_n t + \phi_n)" ) self.play(Write(formula), run_time=3) # Then renders: 1080p60 video, 15 seconds.
Live demo: describe a formula, watch it animate into a video.
