Featured project
Public
Soundpetal
A touchscreen instrument and studio, reborn.
Formerly Maestro Touch · now live at soundpetal.com
The studio opens without an account. Account features require sign-in.

Background
In 2011, as a bootstrapped startup, the original Maestro Touch hit 20,000 downloads and was featured in Apple's App Store in its first week. Its novel interface visually represented relational music theory — but at its core it was a digital instrument. Building it out to its full vision was cost-prohibitive for a passion project in a niche market, so the concept was never fully realized.
Concept
Soundpetal is that idea reborn — a full studio and community, not just an instrument. An AI theory coach watches what you play; sheet music can be read in from a photo; AI visual themes and ElevenLabs sound effects round out the studio. A social Explore feed and shareable pages make creations spreadable, with a quest-and-referral reward system built into the source.
Build story
AI-first building brought the project back to life with a power and vigor that were previously impossible. Built on the orchestration system, its quest implementation checks saved learning, composition, recording or jam activity before completing a quest. The reward code calculates Pro days on the server and records grant references to suppress repeat claims; the SQL denies client roles direct access to the grant-writing function. A deployment ledger records staged backend work, and selected screens use those flags to disable unavailable features. These describe what is built; the ledger still marks some integrations as undeployed.
Outcomes
The public studio opens without an account, and the pricing page publishes plan features and limits. The October 18, 2026 CI run passed 2,424 tests across 145 files. That verifies the source checks, not every hosted workflow: custom instruments, account deletion and Stripe grant synchronization remain marked undeployed in the repository's ledger. No revenue is claimed here. The original instrument has grown into a studio and community — a microcosm of the whole thesis.
Explore the public app
Safety patterns, adapted per app
The “Fleet Kit” describes recurring safety and growth patterns, not a shared package. Each app implements its own version. These examples are visible in the source; they are not a new audit of the hosted services.
In Soundpetal: The music assistant filters instruction-like text and wraps user messages as data; message, token and burst limits bound usage.
- Absorby →Untrusted text is sanitized and fenced before selected AI calls; feature-bucket usage checks compare successful calls with daily limits, without an atomic spending guarantee.
- Glyph →Concept generation sanitizes study text. Its shared limiter uses user, session or IP-based request buckets, but allows requests through if the limiter service fails.
Tech / stack
- React · TypeScript · Tailwind
- Supabase/Postgres · activity checks and reward RPCs
- Stripe · ElevenLabs (some backend work staged)
- AI theory coach · usage limits
- Vitest · GitHub Actions · deployment ledger
The studio opens without an account. Account features require sign-in.
Work with me
Let's talk.
Forward-Deployed / Applied-AI Engineer and enterprise AI-transformation leadership.
Start a conversation ↗More from the fleet
Every app is one click away.
- Absorby
Personalized AI news briefings.
Open →
- Glyph
AI adaptive learning that fits the learner and the subject.
Open →
- Advisor OS
The meeting-to-done copilot for financial advisors.
Open →
- Skill Forge
Coding-interview prep, forged into an adaptive mastery path.
Open →
- Fluidit.ai
AI-generated copy and palette experiments, with Thompson-sampling allocation.
Open →