Problem → Product
When I took on a shop steward role offshore, I needed a better way to help members understand their collective agreement.
People were putting agreement questions into general chatbots and getting answers that were hard to trace back to the book. I wanted the article, the actual wording and an explanation that made sense.
That became STU. Building it uncovered the next problem. Agents needed memory. Projects needed shared computing. Websites needed a better way to get online. Documents needed a home.
One useful tool became an ecosystem I design, build and run myself. Some projects are usable today. Others are still being developed. They're connected by the same habit: find the real problem, build something useful, and keep learning from it.
Problem: Agreement questions needed answers grounded in the agreement itself.
I researched retrieval-augmented generation and built STU around the actual collective agreement: find the relevant article, show its wording, then give a plain explanation. That was how I accidentally got into AI. It's still my favourite project.
My first AI project. Still my favourite.
Problem: Useful context was getting lost between agents and sessions.
The love of STU led to Thaladex: a memory platform for what agents knew, when they knew it and what changed. The aim is durable continuity that agents can use and people can inspect.
Problem: Several products needed the same embedding work, with hardware becoming another thing to manage.
Diffuser gives that shared work a common place to run. A new project should be able to connect to the service rather than solve the same hardware problem again.
Problem: A guide to Newfoundland should feel like it knows the place.
Got creative with this one. Skipper is being shaped as a voice guide to Newfoundland: places, history and stories, with a local voice. “What are ya at.”
Problem: Managing a shared web-edge configuration through agents kept creating friction.
I wanted a clearer interface for adding and maintaining sites. Beacon puts route records behind an API and agent interface, so the work can be expressed through structured operations.
Problem: Documents were drifting between computers, platforms and agent sessions.
Explorer gives rules, runbooks and designs a shared home. People and agents can find the current version and follow its history instead of guessing which copy to trust.
Problem: Running a family Minecraft server became a job of its own.
After unwanted visitors reached the server, I moved it to a whitelist and started building a more useful game master. AUX is intended to help with ordinary server tasks through conversation, such as helping a stuck player or managing a friend's access.
Problem: Managing the tools from a phone needed one understandable place to start.
Cadence OS brings the operator experience into the browser. It gives me a place to reach the tools and see what is happening, while the wider Cadence work develops the persistent chief-of-staff experience.
Problem: Late pipe-spool problems can create rework and another round of verification.
Trammel brings the work back to the shop floor. It is being developed around pipe-spool confirmation and correction, shaped by the practical cost of finding a problem late.