I build tools for the era of private intelligence. Every project in this portfolio starts from the same premise: the person using the compute should be the person in control of the compute. Not the platform. Not the subscription service. Not the company whose terms of service changed last Tuesday.
The work documented here was built in approximately one month in early 2026. It produced two primary inventions that did not exist before this project. The Ghost Engine — a persistent Core 0 intelligence collection process that wardrives, scans BLE, and logs GPS continuously, always, regardless of what the operator is doing, on a $50 device. And the SPI Bus Treaty — the first named architectural standard for shared-bus arbitration on the ESP32-S3, now part of the public technical literature, which is the reason the Ghost Engine never stops. The Ghost Engine is the operational paradigm. The Treaty is the engineering standard that makes it possible. Five additional engineering solutions with no prior documented existence for this platform accompany them, along with a distributed local AI agent framework, two security analysis tools, two open-source publishing tools, and a STRIDE-based threat modeling tool — all deployed, all verifiable against public codebases, all built under the Clark Beddows Protocol: local first, no gatekeepers, you own everything.
The receipt for the embedded OS work is $144. The enterprise equivalent is measurably higher. That gap is the argument.
The Ghost Engine is the operational paradigm: a persistent Core 0 process that wardrives, scans BLE beacons, and logs GPS coordinates continuously — always, regardless of what the operator is doing on Core 1. You can play a game. Use the AI terminal. Browse files. The Ghost Engine never stops. The operator never has to think about it. No previous ESP32-S3 project implemented this architecture. It did not exist before this project.
The SPI Bus Treaty is the engineering standard that makes the Ghost Engine possible. A formal behavioral protocol governing shared-bus arbitration on the ESP32-S3 — the first of its kind, now part of the public technical literature. Without it, the Ghost Engine and the LoRa radio destroy each other over the shared SPI bus. With it, Core 0 runs unconditionally and indefinitely under real-world field conditions.
Built on this foundation: 47 applications, hardware-secured Ghost Partition with sub-millisecond data destruction, local AI inference, LoRa mesh radio, and an ELF module runtime for SD card app deployment without reflashing — on a $50 device.
A custom Debian-based Linux distribution targeting x86 and ARM hardware from 2012–2016. The initial target is the Fujitsu Stylistic Q508 — a 10-inch tablet available for $50–100 on the secondary market — turned into a complete security analysis workstation with native integration of the Pisces Moon OS T-Deck Plus as a dedicated radio coprocessor.
A Python wrapper around Ollama that transforms a local language model into a capable intelligence agent. Before each prompt, The Phantom scrapes the web, bypasses paywalls, pulls live data from official APIs, retrieves semantic memories via ChromaDB, fetches YouTube transcripts, and reads RSS feeds — injecting everything as context before the model sees the query. The model reasons. The Phantom does everything else. No data leaves the machine.
A publicly accessible AI chatbot running on a MicroPC the size of a paperback book, tunneled to the public internet via Cloudflare. Demonstrates that a functional public AI service requires neither AWS infrastructure nor venture capital. Named after Steve Wozniak's 1970s Dial-A-Joke phone line.
A lightweight Python server that sits between any HTTP client and your existing Claude.ai or Gemini browser session. It types your message into the chat window, waits for the response, reads it, and returns it as JSON over a simple REST API. No API key. No per-token billing. No extra subscription. The AI sees a human typing at human speed. Your device gets a JSON response. Your existing subscription — with your Projects, your memory, your conversation history intact — now has a programmable interface. Any HTTP-capable device is a client: ESP32, Raspberry Pi, curl, anything.
A standard API call returns "Hello! How can I help you today?" The Lighthouse returns a continuation of an existing relationship. That is the difference. That is the point.
A native application wrapper that turns any HTML/JS app into a real desktop or mobile application. macOS, Linux, Windows, Android, and iOS — all built and working. The window.spadra.* bridge gives web apps direct filesystem access, serial port communication with hardware including the T-Deck Plus, OS notifications, and app launching. Four apps ship with the framework: a home launcher, a Claude desktop wrapper for Linux, a Ghost Partition SD card manager for Pisces Moon OS, and a Wardrive Splitter for WiGLE CSV files. The deployment layer for the entire Fluid Fortune ecosystem.
A single-file RF intelligence analysis platform for WiGLE-format wardrive datasets. Drop a CSV, receive an interactive heatmap, anomaly detection for mobile hidden networks and evil twin candidates, OUI vendor enrichment, persistent device tracking across sessions, and export-ready reports. No server. No installation. No external API calls except map tiles.
Describe your application's architecture in plain language and receive a structured threat model built on the STRIDE framework — the same methodology used by security teams at Microsoft and Google. Covers all six threat categories with an overall risk assessment. User supplies their own Gemini API key, held in memory only — never stored, never transmitted except to the Gemini endpoint.
A single HTML file blog publishing environment. Write in a word-processor-style interface, publish via the GitHub API directly from the browser. On publish: generates a fully-styled HTML post with Open Graph and Schema.org metadata, updates the post manifest, regenerates the sitemap, and regenerates the RSS feed. Includes a bootstrap utility for migrating existing posts.
A single HTML file podcast publishing environment. Add an episode, enter the archive.org audio URL, publish. Generates the episode page, updates the show homepage with inline players, and regenerates a valid RSS feed compatible with Apple Podcasts and Spotify — all via the GitHub API from the browser. Audio hosted permanently on the Internet Archive. Cannot be deplatformed.
A single HTML file website builder. Choose a theme, pick a layout, add sections, add pages, publish to GitHub Pages via the GitHub API from the browser. Six themes each with complete embedded CSS and matched typography. Eight section types including hero, two-column, cards, CTA, and custom HTML. Multi-page with auto-generated navigation. The third tool in the Fluid Fortune publishing stack.