Building a Sound Machine with AI
Otto Climan documents his ambitious project to design and build a custom ambient sound generator device by combining his own skills with Claude AI assistance. He covers the entire workflow from conceptual brainstorming through 3D design, component sourcing, and software development, revealing both the power and limitations of AI in translating vision to reality.
Why Learning to Build Matters
Building reveals hidden complexity
What appears trivial on the surface often requires weeks of work to solve correctly. Understanding how things actually work teaches you that problems are far more intricate than they initially seem.
Appreciation for design decisions
Every object around you contains countless decisions made by someone. Building teaches you to stop taking others' work for granted and recognize the intentionality behind everyday products.
Discovering your own capability
Building projects reveals that you are capable of far more than you initially believe. The act of creating something tangible expands your sense of what is possible.
The Inspiration and Concept
Three influences converge
Otto combines his love of electronic music (synth-based hypnotic soundscapes), passion for tactile electronics and tinkering, and his use of Claude AI to create something new. The trigger was discovering a sound generator device called Casabrutus in a newspaper.
Sound generator as meditation tool
The Casabrutus device generates ambient sounds for relaxation, but Otto noticed that certain timbres and sound patterns put him into a trance state where he solves problems. He decided to build his own version, scaled up using Claude AI.
Project Requirements and Scope
Five core design requirements
Otto established specific criteria: self-produced (no outsourced help), include a screen for visual focus, incorporate a light sensor for sound modulation, achieve a hi-fi aesthetic suitable for high-end audio equipment placement, and avoid requiring a mobile app.
Multidisciplinary skill requirements
The project demands expertise across design (aesthetics and finishes), electronics (making components function), 3D modeling (prototyping and CAD files), programming (software and firmware), and marketing (selling the final product).
Claude AI Workflow: Design Phase
AI generates navigable 3D browser model
After receiving Otto's detailed brief, Claude created an interactive 3D model that could be viewed in a web browser, which Otto found surprising and useful for visualization before any physical prototyping.
Iterative rendering and refinement
Claude generated multiple renderings based on Otto's verbal feedback, progressively refining the design. Otto rejected initial versions and requested changes like fabric instead of brass, a more modern aesthetic, and an oscilloscope-style screen.
Material redesign for manufacturability
Otto initially wanted a wooden chassis but realized he needed to work with online manufacturing services rather than Italian craftsmen. Claude redesigned the device as a metal (steel) object with wood side panels, making it compatible with Chinese online fabrication services.
Component sourcing by AI
Claude searched online and found specific components including front panel switches, power button, speaker, and checked datasheets to verify power compatibility with Raspberry Pi. However, the actual components differed from the rendered versions.
Reality vs. Rendering: First Problems
AI overestimated design possibilities
Claude had a different understanding of what was physically possible versus what was actually available. The rendered switches looked refined with proper aluminum panel integration, but the actual components were crude and protruded awkwardly.
Power button scale mismatch
Claude placed a power button that was disproportionately large in the design, like a banana among monkeys. Otto had to correct this discrepancy between the rendering and the actual component specifications.
Iterative bouncing to resolve gaps
Otto and Claude went back and forth multiple times to reconcile the design with real component availability. Otto also created a front-view diagram with all components positioned correctly to guide the final assembly.
Software Development with Claude
AI develops testable firmware
Claude wrote software that Otto could test on his computer before deploying to the Raspberry Pi. The software uses keyboard controls (spacebar to start, keys 1-8 for sound layers) to simulate the physical switches and light sensor.
Sound design constraints
Otto specified that all sounds use a pentatonic scale to ensure they sound positive and harmonious together, preventing cacophonies and beat interference. The light sensor modulates randomness within the sound generation.
Multiple firmware iterations and hardware failures
Claude reprogrammed the firmware three times to make it work correctly. During testing, Otto burned two circuit boards, indicating the complexity of translating software to actual hardware.
Project Status and Next Steps
Components ordered, assembly pending
Otto has placed the order for all hardware components, which are expected to arrive within the week. The next episode will cover physical assembly and testing the firmware on the actual Raspberry Pi hardware.
Series continuation planned
Otto is documenting the entire build process in a video series and invites viewers to subscribe to see how the experiment unfolds. He expresses genuine enthusiasm about the project despite its challenges.
Notable quotes
Why don't I try to make one myself from the bottom of my skill set, but scale it up using Claude? — Otto Climan
Claude had a very different idea of possibility than what the reality is. — Otto Climan
Everything has to sound positive, sound good one after the other, so that there are never cacophonies, that there are no beats. — Otto Climan