Robert McCullock
AI Systems Architect & Full-Stack Engineer
I design and ship production AI systems — defining the architecture and constraints, directing AI coding agents to implement, then auditing, testing, and deploying every result. Three multi-agent systems run in daily production; two commercial desktop apps are live for sale; and a free 31-class AI coding academy documents the method. Based in Boston, with a technical career that started at an ISP during the 1990s TCP/IP build-out.
Quick Answer — Who is Robert McCullock?
Robert McCullock is a Boston-based AI systems architect and full-stack engineer. He designs production AI systems, directs AI coding agents to build them against a defined architecture, and personally audits, tests, and deploys every result. Three multi-agent systems run in daily production — an 11-agent publishing orchestrator, a nine-mode creative studio running 6+ months, and a product-design pipeline live since November 2025. He has shipped two commercial desktop apps (VibeTube AI at $49.99, PromptDJ NS-9000 at $79.00), built a free 31-class AI coding academy, and rebuilt 20+ Shopify sections. He is open to full-time roles at Robert@ddsboston.com.
Key Takeaways
- 1Runs three multi-agent systems in daily production. Sovereign Orchestrator Pro (78 files, 11 agents), The Synthetic Director (nine-mode studio, 6+ months live), and NICHE-FORGE-CORE (in production since November 2025).
- 2Ships real, sellable products. Two commercial desktop apps are live and purchasable — VibeTube AI Desktop Pro ($49.99) and PromptDJ NS-9000 ($79.00) — both Windows, Electron, Bring-Your-Own-Key, with serverless DRM.
- 3Full-stack delivery. Production work across React, TypeScript, Shopify Liquid, Electron, Python, and Node.js — data layer to deployed UI.
- 4Hybrid local/cloud AI. Runs local LLM inference on an RTX 3060 through Ollama alongside cloud models (Gemini, Imagen, Veo, Lyria), keeping operating costs low and measurable.
- 5Teaches the method publicly. Designed, built, and deployed the free 31-class DDS Vibe Academy — no paywall, no signup.
- 6Formally trained, deeply experienced. A technical high-school Electronics program (computer networking, service and repair) with straight-A grades, then a career starting at an ISP in the 1990s.
Intent-based engineering
I treat AI coding agents as implementation capacity, not autopilot. The engineering judgment — architecture, constraints, review, and the decision to ship — stays with me. This is how one person ships production systems at the pace of a team.
Define
Architecture, data model, interfaces, and hard constraints — explicit and testable — before any code is written.
Direct
AI coding agents implement against the spec in scoped, reviewable units of work.
Audit
Every output reviewed for correctness, security, accessibility, and adherence to spec before it is accepted.
Test
Behavior, edge cases, and performance validated. Structured data and Core Web Vitals confirmed where they apply.
Deploy
Shipped to production, verified live, monitored. Iteration driven by real signals, not assumptions.
Three systems in daily production
These are not demos. Each of the three is a multi-agent system running real work every day — verified against the live codebases. File counts, line counts, model rosters, and run dates below are drawn from forensic audits of the source.
An autonomous publishing engine that researches, drafts, compliance-checks, and posts content across WordPress, Shopify, X, and Reddit from a single execution queue. It coordinates an 11-agent pipeline — ten named Gemini-backed agents plus a local Ollama cluster — running 19 concurrent threads. A five-cycle local intelligence engine sequences five open-source models on the RTX 3060 overnight to analyze performance data at near-zero cost, while a strict budget service and exponential-backoff layer keep the system inside hard daily API limits.
My role: Full system architecture — the agent pipeline, the local/cloud model routing, the thread scheduler, the budget and rate-limit safeguards, and the React dashboard. Verified daily run cost is roughly $8–$13 from the system's own cost telemetry.
A desktop-first creative studio that automates the full content pipeline — ideation, scripting, image generation, and cinematic video. Its interface is a nine-mode workstation (influencer factory, social command, blog engine, trends radar, video studio, storefront integration, creative protocol, generator forge, telemetry) rather than a chat box. A tiered model cascade routes work across Google Gemini, Imagen, and Veo by cost and quality, with local PostgreSQL persistence. It has generated the daily social content for Design Delight Studio for over six months — a real, sustained production track record.
My role: Full architecture — the nine-mode state machine, the model-tiering cascade, the local persistence layer, and the React 19 desktop UI. Real-time cost telemetry is built in so every API call is measured.
A browser-based application that automates the full product-design pipeline for a print-on-demand brand — chaining multiple Gemini models through a six-step workflow: live web trend research, creative strategy, image rendering, SEO metadata, and direct Printful draft creation. State persists locally through IndexedDB; a lightweight Express proxy secures the Printful credentials. Production session artifacts confirm continuous commercial use since November 2025. The real intellectual property is the prompt-engineering layer — a tuned brand bible, 30 creative archetypes, and a large mutation pool.
My role: Full build — the six-step pipeline, the multi-model routing, the IndexedDB persistence, the Printful integration and proxy, and the prompt-engineering IP. Built and run as a single-operator production tool.
How a multi-agent system fits together
A simplified view of the Sovereign Orchestrator pipeline: a research scout feeds a strategy agent, a writer-and-reviewer loop produces compliant content, and a thread scheduler publishes to four platforms — while a local model cluster learns from results overnight.
Simplified pipeline view · Sovereign Orchestrator Pro V6.0 · 11 agents, 19 threads, hybrid local/cloud inference
Two apps shipped and selling
Designing software is one thing; shipping it as a packaged, licensed product someone can buy is another. Both of these are live in the Design Delight Studio storefront.
A commercial Windows desktop app that turns a single URL into a complete set of YouTube content assets — titles, descriptions, tags, script outlines, thumbnail prompts, and cross-platform posts. Bring-Your-Own-Key, so the buyer pays no recurring fees; keys stored in the OS keychain; license verified through serverless DRM.
My role: Full product — Electron shell, React UI, generation pipeline, BYOK key handling, DRM backend, installer, Shopify product and landing pages.
View product ›A commercial Windows desktop app that synthesizes continuous live music from text prompts and EQ controls, using Google's Lyria realtime model over a streaming session. A custom AudioWorklet encoder exports studio-grade WAV files directly to disk — impractical in a browser sandbox, which is why it ships native. 40 tuned style presets, BYOK.
My role: Full product — the real-time audio engine and EQ chain, the Lyria streaming integration, the Lit web-component UI, key vaulting, DRM, and Electron packaging.
View product ›| Detail | VibeTube AI Desktop Pro | PromptDJ NS-9000 |
|---|---|---|
| Price | $49.99 one-time | $79.00 one-time |
| Stack | Electron, React, Vite | Electron, Lit, Vite |
| AI model | Google Gemini (BYOK) | Google Lyria realtime (BYOK) |
| Licensing | Serverless DRM, Cloud Run | Serverless DRM, Cloud Run |
| Recurring cost to buyer | None | None |
The DDS Vibe Academy
I designed, built, and deployed a free 31-class curriculum on AI-assisted software development — covering Claude Code, Google Antigravity, Gemini, Cursor, Shopify Sidekick, and Ollama. It is a tabbed 12-section Shopify hub with an interactive knowledge map and full structured-data coverage. No paywall, no signup, no certificate. It demonstrates something a hiring manager should care about: I can not only build production systems, I can document and teach the engineering method clearly enough for others to follow.
Visit the AcademyEverything built
The complete set of systems and applications. Tagged by current state: Shipped means live, in use, or for sale; Prototype means built and working but not yet released as a finished product.
Autonomous publishing across four platforms. 78 files, ~21,700 lines, 11 agents, 19 threads. In daily production.
Automated video and image production. ~38 files, ~13,700 lines. Live 6+ months.
Six-step POD design pipeline. 24 files, ~3,700 lines. In production since Nov 2025.
URL-to-YouTube content generator. Electron + React, BYOK, serverless DRM.
Generative audio workstation. Electron + Lit, Google Lyria, studio-grade WAV export.
12-section Shopify learning hub, ~6,400 lines, interactive knowledge map.
Automated content pipeline for WordPress with a strategy layer. 14 files, ~1,650 lines.
A match-3 browser game with a self-healing board. Live and playable.
A turn-based strategy game with an NPC personality model. Landing page live.
A comprehensive crochet stitch reference, published as a live interactive page.
The custom Shopify theme behind ddsboston.com — section library, structured-data system, performance baseline.
Bilingual contractor website with booking and form integrations, deployed on Cloud Run.
An AI-powered virtual fitting room that lets shoppers see apparel on themselves from an uploaded photo or live camera. In production 6+ months, embedded on every clothing product page on the storefront.
View live feature ›An internal R&D workbench for multi-modal model experiments.
An adversarial-review tool that stress-tests documents and arguments from multiple angles.
A product-strategy assistant that turns store and market data into recommendations.
A multi-model pipeline for generating and directing commercial product imagery.
A brand-storytelling generator for short-form narrative content.
Shopify & front-end work
Production storefront engineering on ddsboston.com — measurable front-end work that maps directly to a product or e-commerce engineering role.
Rebuilt 20+ storefront sections to Online Store 2.0 standards with schema-driven settings, scoped CSS, and WCAG 2.1 AA accessibility.
A measured performance pass that materially reduced largest contentful paint and total blocking time and lifted the mobile performance score, with platform ceilings documented honestly.
A reusable structured-data system — multiple JSON-LD schema types per page template — for rich results and AI-search visibility. This page uses it.
Custom page templates including the storefront's about, contact, policy, FAQ, and landing pages — each with full semantic structure.
Participant in the Microsoft Clarity pilot program, using session analytics to inform storefront UX decisions.
Implemented analytics and tag management across the theme, with event tracking wired to real storefront interactions.
Technical skills
Grouped by area. These are tools used in shipped or running systems — not a checklist.
- TypeScript
- JavaScript
- Python
- Liquid
- HTML5
- CSS3
- SQL
- React 19
- Lit
- Vite
- Tailwind
- Web Audio API
- Web Components
- SVG/Canvas
- Multi-agent design
- Prompt engineering
- RAG
- Ollama
- Gemini
- Imagen / Veo
- Lyria
- Electron
- Node.js
- Shopify OS 2.0
- WordPress
- Google Cloud Run
- Firebase
- PostgreSQL
- IndexedDB
- GraphQL
- localStorage
- OS keychain
- Code audit
- Core Web Vitals
- JSON-LD / SEO
- WCAG 2.1 AA
- DRM / licensing
- Local-first design
- Claude Code
- Google Antigravity
- Cursor
- Git
- electron-builder
- Local LLM hosting
- RTX 3060 / CUDA
- Serverless DRM
- OTA updates
- Networking
Training & career
Formal technical training, then a career that began before the consumer web and continues through to production AI systems.
Electronics — four years, straight-A grades
Completed a full technical high-school program in Electronics, including computer networking and equipment service and repair. Earned straight-A grades across all four years — the formal foundation under everything that followed.
Internet Service Provider — early technical role
Began his professional technical career at an ISP during the TCP/IP build-out of the early commercial internet. Hands-on networking and systems work.
Founded Design Delight Studio
Launched a Boston sustainable-apparel brand and took on all of its engineering — storefront, theme, and infrastructure.
Storefront engineering & first AI systems
Built the storefront's section library and structured-data system, ran the Core Web Vitals sprint, and brought the first multi-agent systems into production — NICHE-FORGE-CORE live from November.
Daily production systems, commercial apps & the Academy
Scaled the orchestrator and creative studio to daily production use, shipped VibeTube AI and PromptDJ NS-9000 as commercial desktop products, delivered a freelance client website, and designed, built, and deployed the free 31-class DDS Vibe Academy.
Where I fit
The kinds of teams and problems where the work above translates directly.
AI product teams
Teams shipping AI features who need someone who has taken multi-agent systems from architecture to daily production — and audited them.
Full-stack / front-end teams
React and TypeScript teams that treat accessibility, Core Web Vitals, and structured data as engineering concerns, not afterthoughts.
E-commerce engineering
Shopify and storefront teams that need someone fluent in Liquid, theme architecture, performance, and SEO.
Desktop app teams
Teams building Electron applications — including licensing, DRM, secure key handling, and auto-update.
Developer tools & education
Companies in the AI developer-tools space who need someone who can both build the product and teach the workflow clearly.
Lean / founding teams
Early teams that need one engineer who can own a surface end to end — architecture, build, ship, and iterate.
Frequently asked
Bottom Line
Robert McCullock is a Boston-based AI systems architect and full-stack engineer who ships real, working software: three multi-agent systems in daily production, two commercial desktop apps live for sale, a free 31-class AI coding academy, and measurable Shopify storefront engineering. He builds by defining the architecture, directing AI coding agents to implement, and auditing every result before it ships — production systems at the pace of a team, without giving up engineering judgment. Formally trained in Electronics and working in technology since the 1990s, he is open to full-time roles and can be reached at Robert@ddsboston.com.
The brand I build for
Design Delight Studio is the Boston sustainable-apparel company I founded and engineer. Its products carry five independent sustainability certifications — context for the standards I hold my own work to.