Projects
Self-hosted AI systems, embedded hardware and production WordPress plugins — designed, built and run end-to-end on my own infrastructure, from ESP32 firmware to autonomous multi-agent pipelines.
3 live on this site
24/7 autonomous agents
100% local-first AI
Live on this site
Three finished systems you can explore right now — each one runs in production here, powered by hardware and models I host myself.
Satellite Dashboard
A DIY GPS telemetry pipeline: an ESP32‑S3 with a u‑blox NEO‑M9N receiver drives ten on‑device screens, parses raw UBX and NMEA, then streams JSON over Wi‑Fi to a WordPress dashboard — live track maps, satellite sky plots, signal charts and even jamming/spoofing integrity flags.
Explore the live dashboardAI Book Writer, Editor & QA
An autonomous publishing house inside WordPress: a planner outlines, a writer publishes pages around the clock, a continuity editor keeps a “story bible” and applies reversible fixes, and a 16‑agent QA pipeline gates and scores every book out of 100 — all on local Ollama models, with AI cover art and EPUB/PDF exports.
Browse the books it wroteZumba Routine Generator
AI choreography for real instructors: type a song and get an editable, colour‑coded, print‑ready cue sheet phrased in 32‑count blocks — grounded in the track’s true BPM and energy via Spotify (with a Deezer fallback), with save/load, CSV export and AI poster art.
Generate a routineMore projects
Decision engines, retrieval systems, always-on agents and home IoT — the connecting thread is that every AI feature runs on hardware I own.
AI Investment Engine
Explainable BUY / WAIT / SELL decision support — live indicators, forecasts and AI‑scored news behind hard risk gates the AI can’t override, with a Telegram kill‑switch.
Local RAG System
A from‑scratch retrieval engine: boundary‑aware chunking, ChromaDB vectors and cited answers that refuse to guess — with CI, Docker and a full audit log.
LlamaIndex RAG
The same “ask your documents privately” idea rebuilt on the LlamaIndex framework — a clean comparison of hand‑rolled vs framework retrieval.
VPS RAG Chatbot
This site’s own chat widget: retrieval‑augmented answers over my docs from a VPS‑hosted LLM, behind three independent CPU and memory guardrails.
News Story Weaver
Scans live news feeds, extracts verified facts and weaves them into genre‑fiction briefs for the Book Writer’s queue — later absorbed into the flagship plugin.
Document Impact Agent
An always‑on Windows service that assesses engineering document changes — service bulletins, CMMs, drawings — and writes evidence‑cited impact reports, plus its own test harness.
Home AI Assistant
A private, phone‑friendly assistant on the home network: chat, shared lists and dated reminders on local Ollama — password‑gated, no cloud, runs as a service.
ESPHome Bluetooth Proxy
A £5 ESP32 flashed as an ESPHome Bluetooth proxy, relaying BLE to Home Assistant over Wi‑Fi and extending smart‑home range across the house.
Local-first by design
Every AI feature on this page runs on hardware I own — a GPU desktop for generation, a VPS for serving, ESP32s at the edges. No data leaves the network, no per‑token bills, and no third‑party model ever sees a word. The books write themselves overnight; the electricity bill is the only subscription.