Developer Relations · Technical Content · Community

I build AI developer tools —
and the story around them.

Shipping Claudot, a Claude Code integration for the Godot engine, taught me the whole job: the MCP tools, the bridge daemon, the docs, the trailer, the brand.

Fifteen years of writing and producing before that — off-Broadway theater, then League of Legends content seen by hundreds of millions of players. Now I point all of it at helping developers get real work done with AI.

Read the résumé Browse the code

Featured

Claudot

Solo creator

2025–present · Source-available · Python + GDScript · Godot 4.2+

Claude Code, docked inside the Godot editor. Chat with an AI that can actually see your project: inspect the scene tree, modify nodes, run the game, capture screenshots, and read the debugger — without ever leaving the editor.

  • Authored 20 custom MCP tools covering scene inspection and modification, screenshot capture, and GUT test execution — so Claude can write code, run it, see the result, and iterate. It doubles as a working reference for extending Claude Code into a new developer environment.
  • Shipped a Python↔GDScript bridge daemon and a dockable editor chat panel with slash commands, @ file references, and persistent history.
  • Multi-provider by design: Claude Code and the OpenAI Codex CLI, plus bring-your-own-key access to the Anthropic API, OpenAI, OpenRouter, and any OpenAI-compatible endpoint (Ollama, LM Studio).
  • Owned the entire developer experience end to end — zero-terminal install, onboarding flow, documentation, sample workflows, UX copy, marketing site, launch trailer, and visual identity.
  • Python
  • GDScript
  • MCP
  • Claude Code
  • Godot 4
  • Daemon / IPC
  • Editor plugin
  • GUT
claudotcode.com Source 7 stars

Selected work

BLANK the spire

C# · Python · Flask · Slay the Spire 2

Forge a playable Slay the Spire 2 character class from a single sentence. Three pieces working together: a data-driven C# mod runtime (Harmony + the community SDK) that turns a class defined as pure data into a real character, a multi-stage LLM generation pipeline that writes cards, relics, and mechanics, and a Flask site that ships it to players.

The interesting constraint: generated content has to be guaranteed playable. A published card/relic/status vocabulary bounds what the model may invent, and schema validation rejects anything the runtime can't execute.

  • C#
  • Harmony
  • Python
  • LLM pipeline
  • Schema validation
  • Flask
Live site Source

Holmes

Private

Python · spaCy · LLM orchestration

Point it at a pile of text, name a character, and get back a structured bundle for teaching an LLM to play them: dialogue pairs, memories, a voice profile, relationships, reputation, and chat-format JSONL ready for training.

Format-agnostic on both axes — PDF, DOCX, plain text, JSON going in, and novels, screenplays, historical documents as the kind of thing it reads. The design principle is the cost story: code reads everything; the model reads only what requires comprehension. Parsing, quote extraction, and vocabulary statistics are deterministic and free; the paid stage touches only the scenes the character appears in, once.

  • Python 3.11
  • spaCy
  • NLP
  • OCR
  • Training data
  • CLI

Private repo — happy to walk through the architecture.

Trainything

In progress

Human-in-the-loop tooling

Most teams know they need humans in their training and feedback loops. What stops them is the plumbing: wiring up review queues, collecting judgments in a usable shape, and getting that signal back into the model.

Trainything is the tool I want to exist for that — get humans into the loop as quickly and effectively as possible, so the bottleneck goes back to being the quality of the judgment rather than the cost of collecting it.

  • Evals
  • RLHF / feedback loops
  • Annotation UX

Early stage — in active development.

Background

I came to developer tooling the long way around. Before I was writing MCP servers I was writing scripts — for theater, then for video, then for games.

That turns out to matter more than I expected. The hard part of developer relations isn't the demo, it's the explanation: figuring out what a developer already knows, where they'll get stuck, and what order the ideas need to arrive in. I spent fifteen years learning that in front of live audiences and at League of Legends scale, and now I get to apply it to tools I also build.

These days I work in an agentic coding workflow of my own design, every day — which means I have hands-on, unsentimental opinions about where Claude Code shines and where developers actually get stuck.

Based in San Francisco. Organized a GDC meetup on AI in games.

  • 2024–now Story GamesCreative Director — generative AI practice, prompt-engineering standards, character LoRAs, ElevenLabs partnership
  • 2019–2024 Riot GamesWriter & Content Producer — League of Legends video seen by hundreds of millions; shipped Tellstones: King's Gambit
  • 2010–2019 The Puppet Shakespeare PlayersFounder & Creative Lead — nine-year off-Broadway run; Puppet Titus Andronicus was a New York Times Critics' Pick