Master automotive technician (EV and Petrol) and technical trainer specializing in vehicle diagnostics, high-voltage systems, and skilled-trades pedagogy. Tesla alumni with additional non-vehicle exposure and practice in industrial automation, PLCs, power transmission, leadership, and more. Backed by a decade of industry experience and an automotive technology degree, I bring the diagnostic mindset — reproduce, isolate, verify — to hard, tool-agnostic problems.

My projects involve AI in practice: MCP servers, Claude Code workflows, CAD design and prototyping, and agentic authoring pipelines that turn subject-matter expertise into decks, docs, VR simulations, and diagnostic tools that teams I support can actually use. Public work here is the tip of that iceberg.

• claude-bridge-kit — a portable installer for claude-bridge, an MCP server that dispatches the headless Claude Code CLI (claude -p) from inside Cursor. Turns "Cursor chat" into an orchestrator that can hand long-running, autonomous work off to Claude Code in isolated git worktrees, with self-review passes, effort tiers, and preset workflows.
• content-studio — a frontier-LLM-paired content-authoring product for technical training teams, powered by CAA (Course Authoring Assistant). Six deterministic quality gates, a 21-archetype slide surface, Bloom-level discipline, and .pptx output. (private — flip pending; visible on request)
• motor-starter-xr — agentically-authored VR training for diagnosing a 3-phase 480V across-the-line motor starter, built to prove Cursor + the official Unity MCP server can author production-quality industrial training with an on-device Sentis adaptive coach. (private — flip pending)
• pcm-notes — curated public window into PCM, a private multi-agent lab. Charter amendments, fleet architecture, methodology notes, and multi-model evaluations.

• Several small CAD projects and 3D-printed prototypes end to end (originals, not reprints of other people's work)
• Hand-built electrical training aids and harnesses
• Classroom desktop power-supply prototype
• CAN bus simulator w/ Arduinos
• Moderately impressive desktop PC

• Design for the UX — my personal philosophy when innovating anything: Function precedes Form.
• Guardrails first. Tests, isolation, E-stops, DECISIONS.md discipline. If a claim ships, a test proves it; if an agent has a permission, a test enforces it stays scoped.
• Domain depth applied to AI. I bring the diagnostic technician's problem-solving abilities with a technical educator's twist — reproduce, isolate, iterate, verify, explain — to how I build with LLMs. Every agent I ship is scoped by a charter, gated by explicit quality checks, and paired with a human-in-the-loop model matched to the stakes.
• Cross-medium output. Same domain knowledge, multiple output shapes: HTML/.pptx decks, .docx, VR simulations, native mobile companions. The medium is chosen by what the learner needs, not what the tool makes easy.

Claude Code · Cursor · MCP · LangGraph · Python (FastAPI, pydantic, python-pptx) · Node.js · Unity (C#, Sentis) · Swift/SwiftUI · Azure (Container Apps, Key Vault, Blob) · GitHub Actions (OIDC) · Autodesk · Dassault Systèmes

Open to conversations about applied-AI roles where domain depth matters. Reach me through GitHub or LinkedIn.