- What I’m building
- A product that helps people describe an app, shape a plan, build a first version, and make changes as they try it.
- Current stage
- Opening Shinkiro to early design partners. The public site explains the pilot and its limits.
Projects
Products, open source, and experiments.
What I’m building now, the code I share, and earlier experiments. For production AI and organizational work, visit my personal site.
- What it is
- An unofficial Ruby port of Google’s Agent Development Kit for AI agent development.
- Scope
- An independent port, not an official Google SDK. See the repository for supported features and current status.
- What I’m building
- A home for dishes worth remembering, from restaurant discoveries to your own kitchen. Save photos, notes, and what you would order again.
- Current stage
- A web beta with dish logging and shareable Food Folios. New dishes start private; ranking every meal is not required.
- The problem
- Every RAG tutorial started with a vector database, a framework, and eleven dependencies — so Rubyists reading them concluded this wasn't for them.
- What I built
- The whole RAG architecture in one small gem: no vector database, a few light dependencies, and an API key. Indexing, embedding, retrieval, generation — readable in one sitting.
- Purpose
- A compact reference for understanding indexing, embedding, retrieval, and generation in Ruby.
Open source
Live
Ruby ReAct Agent
reasoning and acting, no framework required
Ruby Agents Tool use GitHub ↗- The problem
- Agent frameworks hide the loop, which is exactly the part you need to understand before you trust one in production.
- What I built
- A direct implementation of the ReAct pattern in Ruby — reason, act, observe — with real access to external tools and nothing clever in the way.
- What changed
- A reference people can read to see where autonomy actually needs brakes.
- The problem
- Python had Streamlit. Ruby had nothing for standing up a quick data app around a model.
- What I built
- A Streamlit-like framework for Rubyists — declarative UI over a script, so an experiment could become something you show someone.
- What I learned
- Tooling gaps are community problems, not language problems. Also: shipping a framework is a lifestyle, not a weekend.
- The problem
- AI coursework lives in notebooks, and notebooks meant Python — so Rubyists were locked out of the standard way to learn this material.
- What I built
- IPython magic that runs Ruby inside Google Colab, plus the writeup that shows how to get there in a few cells.
- What changed
- Ruby developers can follow along with an ML curriculum in the language they think in.
Nothing in that group.
try another filter.
Earlier projects
older, smaller, still mine- Forecasting the Future Materials from the RailsConf talk: a full machine-learning project in native Ruby, using linear regression because it explains itself. GitHub ↗
- BannerCreator.pro An earlier experiment in creating banners and social headers.
- Expand.ly An earlier experiment in AI-assisted writing and learning.
The architecture I keep drawing
Have a decision you’re working through?
Explore advisory work
I’m open to selected advisory engagements on AI strategy, product direction, and adoption.