Software Engineer • AI Enabler • Creative Technologist
"Coming to a galaxy near you!"
// transmission incoming
What Is AI JEDI?
Mike is a software engineer who doesn't just talk about AI — he ships with it.
Every project on this site was built faster, smarter, and leaner because AI was in the
loop at every stage. From Claude Code writing production systems to Veo generating
film assets, this is what happens when you stop experimenting and start engineering.
The Stooge Network
Four machines. One distributed brain. Running autonomously 24/7 in a spare bedroom.
MoSource of TruthAlways-on server. API hub. The Vault.
CurlyPrimary Dev RigWhere code ships. Debate bot worker.
ShempHeavy ComputeRTX 3070. AI inference. Movie production.
What They're Building
Autonomous AI debate bots — multiple Claude instances arguing positions in real time
Remote admin via HTTP API — no SSH, no file shares, just clean REST from a phone
Telegram command interface — control the whole network from anywhere
AI-generated film production — Veo video, Gemini images, DaVinci Resolve, zero crew
Distributed task execution — watchers on every machine, always polling, always working
This site — designed, generated, and deployed entirely with Claude Code
Jedi Studio
A full AI media production suite running on local hardware. From YouTube research to final render — no cloud editing tools, no subscriptions, no crew.
YouTube intelligence pipeline — scrape channels, extract topics, companies, predictions, and business opportunities into a searchable SQLite knowledge base
Natural language Q&A — ask questions against the knowledge base using Claude CLI, Gemini, or Perplexity
AI script generation — Claude researches the topic, Perplexity writes the script, auto-timed to video clip durations
Image generation — Gemini 3 Pro, Higgsfield Nano Banana Pro, with text overlay compositing via PIL
Video generation — Google Veo 3.1, Kling 2.6, and Higgsfield via Playwright browser automation
Text-to-speech — three engines: XTTS on local GPU (Shemp), Bark neural voices, Microsoft Edge TTS
DaVinci Resolve automation — Python scripting API assembles timelines with video, voiceover, text overlays (Fusion TextPlus), background music, and sound effects
AI director review — automated quality scoring of images and videos with prompt revision suggestions
Drag-and-drop scene reordering, per-scene disable toggles, and full render queue management
The Dashboard
One web UI to run everything. Eight tabs. 300+ API endpoints across the Stooges server and Jedi Studio. Accessible from anywhere via Cloudflare tunnel with zero-trust auth.
Task management — create, assign, and track tasks across all four machines with Jira sync
Full email client — Gmail IMAP/SMTP with AI-powered classification, auto-archive rules, and bulk actions
Remote terminal — execute PowerShell on any network machine via SSH or API fallback
API usage dashboard — real-time cost tracking for Claude, Gemini, Codex, Perplexity, and Higgsfield with Chart.js visualizations
Codex integration — OpenAI code generation, test generation, and documentation with per-model usage metering
Local LLM inference — Ollama on Shemp's RTX 3070 running Qwen3 8B at 67 tokens/sec
Smart home control — Philips Hue bridge pairing, light discovery, and per-light/group control
Business quote system — vendor requests, image uploads, approval workflows
SMS via Twilio — two-way messaging with inbound webhook handling
Infrastructure Stack
Four Windows machines, a Docker-powered VPS, and Portainer orchestration — a hybrid on-prem/cloud stack built for control, speed, and zero monthly surprises.
Docker + Portainer — containerized services on IONOS VPS with web-based management and one-click deploys
Cloudflare tunnel with dual-connector failover — Mo + Shemp both run cloudflared, CF auto-routes around failures
Zero-trust access — Cloudflare Access protects all endpoints, no ports exposed to the internet
Telegram bot — two-way command interface for controlling the entire network from anywhere
The network hasn't stopped shipping. Recent arrivals:
Mary Ann — the secretary-agent that routes every request to the right Stooge on the right machine with the right provider. She classifies intent, resolves agent + machine + LLM, embeds task JSON in chat replies, and never asks permission twice
12 advisor personas — Einstein, Aristotle, Munger, Buffett, Marcus Aurelius, Kahneman, Sun Tzu, Drucker, Shakespeare, Jobs, Darwin, Meadows. Gemini-backed, persona-driven reasoning for decision framing and debate consultation
Galaxy cockpit — 3D Three.js command center with seven planets (one per Stooge plus Mike at the sun), dynamically-rendered agent moons, five camera modes including WASD flight with thrust ramping, and an alien-invasion pentest visualizer with laser combat
Explorable Canvas — a CYOA interactive story engine with physical-book-flip logbook (StPageFlip), layered-image scenes, an E-key layer editor with drag/rotate/skew/3D-perspective, animated sky layers, and a 40-test Playwright stability suite
Persistent memory — SessionStart hooks inject network state into every new Claude session, RAG writeback loops pull learnings back into per-agent knowledge, semantic task cache returns cached results at 0.9+ similarity for zero tokens spent
Agent-oriented skill triggers — a Tier-1 standard skill library universally injected into agent prompts (task-report, research-report, email-format, status-brief) so outputs chain reliably across a whole network of agents
Per-intent turn budgets — the routing engine assigns max-turns contextually (research-live 5, security-audit 25, infrastructure 20) so Perplexity lookups don't waste budget and investigations don't starve
The Bible — 30+ self-published chapters documenting every subsystem, served through a vintage book-flip viewer, with a four-tier operator catch-up on top so any new session boots with full context in ~3K tokens
Integration seams mapped — twelve named cross-subsystem seams (dashboard↔server, server↔RAG, email↔tasks, scheduler↔queue↔watchers, two-brain coordination, tunnel↔Tailscale failover) documented with failure modes and runbooks