AI automation operator proof across systems, media, and music tech
A categorized proof hub for practical AI workflow work: recurring systems, agent operations, research intelligence, creative automation, music/audio technology, and public launch media.
The through-line is not “AI demos.” It is operating judgment: choosing useful inputs, cutting noise, building reviewable outputs, documenting the handoff, and keeping sensitive actions behind human approval.
Practical AI implementation, automation ops, and human-in-the-loop systems
I build approval-gated AI workflows for research, content, marketing, inbox, site QA, and operations work. The best proof is the operating layer: clean-room demos, recurring report systems, safety boundaries, documentation, and creative systems that turn messy inputs into something a person can actually use.
Autolabel by PREBALLIN: a full artist operating system
My most ambitious build: an end-to-end web app that gives independent musicians label-grade structure. It finds an artist across every major platform, builds a live Artist Snapshot under a strict no-fake-stats policy, generates AI strategy, and keeps automation approval-gated. Real integrations, real payments, real data integrity.
Live flagship productAUTOLABEL: ARTIST OPERATING SYSTEMDiscovery-first onboarding, a unified Artist Snapshot, a connector hub, AI strategy, and approval-gated automation, built for independent artists who need a label's structure before they have a label's team.Open live app →
Inside the build Four surfaces of the same system: the Artist Snapshot web, the deep read, the connections hub, and the label desk. Tap any panel to open the live app.
Artist Snapshot · the webA signing-grade read, mappedThe AI writes a full label read covering core themes, visual world, content voice, fan archetypes, influences, and the fit, then renders it as an explorable web you drag through and tap to read.Artist Snapshot · the readLong-form, in the artist's voiceEvery section expands into a written read grounded in confirmed profiles and the artist's own words: no invented stats, just a real bio and strategy the artist can actually use.Connections hubEverything a label asks for, oncePress assets, EPKs, discography and ISRCs, split sheets, and PRO registration: the label-grade paperwork collected in one honest place, with nothing required to start.The label deskYour whole team, in syncOne dashboard that reads from a single source on the artist: ask the room, or hand a job to the right desk, such as the release strategist or content director, with approval-gated automation.
Integrations
Everything wired in
Music platforms: Spotify Web API, Apple Music (iTunes), Deezer, YouTube Data API, and SoundCloud, with live artist and catalog search
Social: Instagram Graph API for official profile and insights, plus X, Twitch, TikTok, Threads, and Facebook search
AI models: GLM-5.2 via Z.ai (OpenAI-compatible), bring-your-own-key for Anthropic / OpenAI / OpenRouter / Gemini, or your own AI subscription over an OAuth runtime
Payments: Stripe subscriptions and Label Balance top-ups, signed webhooks crediting a tier-aware ledger
Email: Resend on a verified domain for artist confirmations
Notifications: Discord webhooks on every intake and payment
Auth and data: Google OAuth with HMAC sessions, Postgres (Neon/Supabase), and Supabase blob storage
Manual analytics: paste and upload layer for Spotify for Artists, Apple for Artists, Chartmetric, Songstats, Soundcharts, and Viberate
How it works
Under the hood
Stack: Next.js App Router on Vercel serverless with a tenant-scoped multi-workspace data model and fail-closed auth
Connector library: normalizes 8+ platforms into one honest schema, showing "unavailable" wherever a platform does not expose data instead of guessing
Artist Snapshot engine: rule-based, never fabricates listeners or followers, and recommends a service tier with reasons
AI layer: generates strategy reads grounded only in confirmed data, with no invented numbers
Agentic but safe: drafts captions, campaigns, and posts automatically while publishing stays human-approved
Billing: Stripe Checkout (no card data touches the app), a signed webhook, then a Label Balance ledger with 20%-markup and at-cost tier logic
Security: AES-encrypted credential storage for bring-your-own-key, with secrets never serialized
Quality: 77 automated tests plus regression guards against fake-stat and branding drift
Featured systems
Start here: strongest operator proof
These are the clearest stories for an AI automation operator pitch: systems with repeatable inputs, useful outputs, safety boundaries, and real operating judgment.
AI operations command center
Hermes automation fleet
Operated a 40+ job automation layer with recurring scans, delivery routing, health checks, archive sync, site QA, lead-intelligence reports, reminders, and approval boundaries.
Proves: recurring AI ops, monitoring, reliability, handoffs
Public-safe status: aggregate story only; raw prompts and IDs stay private
Agent cost + reliability
Agent wake/cost optimization
Turned noisy agent activity into a tighter operating loop by separating always-on checks from true wake conditions, while preserving useful coverage.
Proves: cost control, signal filtering, systems thinking
Public-safe status: verify exact before/after metric before publishing a number
Music education product
Notespawn: Sight-Reading Generator
Built a web app that generates fresh grand-staff piano sight-reading exercises with difficulty, key, time-signature, hands, seed, audio playback, PDF, MusicXML, and MIDI controls.
Public-safe status: use synthetic or already-public inputs only
Interface screenshots
Tools with real screens and system diagrams
Product screenshots show the built interfaces. Clean diagrams explain the private automation systems without exposing logs, credentials, personal data, or internal IDs.
Theater practice appScenePartner / ScenePal rehearsal modeLine-practice interface with hidden lines, character roles, progress tracking, and reveal controls for running a scene.Open live tool →Script parsing workflowScenePartner script import and setupScript import and scene-selection flow that turns messy script input into practice-ready material.Open live tool →Music education appNotespawn: Sight-Reading GeneratorGenerated grand-staff exercise with playback plus PDF, MusicXML, and MIDI export.Open live tool →Creator commerceBeat marketplace playerMusic-commerce interface with playable previews, license tiers, checkout buttons, and a compact audio player.Open live storefront →System architectureHermes automation fleetClean architecture view of recurring checks, report generation, approval gates, delivery routes, and archive/search handoff.Reliability modelAgent wake/cost optimizationBefore/after model for reducing noisy agent runs while preserving useful coverage and observability.Research intelligenceDaily intelligence and triage enginePublic-safe architecture for turning source noise into ranked findings, reject reasons, confidence notes, and next actions.Safe deployment laneApproval-gated site operationsSafe-lane deployment flow that separates draft work from public changes and verifies production before calling work complete.Product proof pipelineProduct interface proof pipelineHow music and rehearsal tools turn structured or messy inputs into playable, exportable, public-facing interfaces.Approval-gated workflowJob and lead workflow abstractionPublic-safe diagram of a human-reviewed pipeline from fit check to proof routing and follow-up draft.
Public proof links
Creative and launch media already visible online
These stay as public proof, but the overhaul now routes them under the larger AI automation operator story instead of letting the whole portfolio read as only a media link grid. The launch media case study adds context on the creative constraints, process, and public work.
Featured creator collab proofRAY J COLLAB - RED NOSESThe large four-frame showcase stays up top as the hero proof piece, followed by the cleaner X/Twitter embed screenshot cards below.Open proof link →
The public media proofs below use static screenshots captured from rendered X/Twitter embeds, so the cards show the public post context plus the video starting frame without depending on live X widgets at page load.
Content radar, landing-page audits, lead research, and follow-up prep for reviewable next actions.
Research intelligence
Digest and triage systems that rank public/source-safe findings with reject reasons and confidence notes.
Safety boundaries
Public pages show clean-room demos, aggregate diagrams, and public posts, not logs, prompts, account data, or secrets.
Public-safe boundary
What is not shown here
Some of the strongest source material involves private logs, personal workflows, credentials, account sessions, or sensitive research areas. Those do not get copied into public pages. When useful, they become clean-room demos, aggregate metrics, diagrams, and synthetic examples.
