Practical AI help for organizations that have to get it right.
Training, adoption standards, and software development for public agencies and small businesses in Astoria, Clatsop County, and the Pacific Northwest.
Decades of writing software by hand, and now the tools to do more of it for less. In the past year: more than ten applications, most of them first written by hand and then rebuilt or extended with AI help, plus a production web app with AI at its core and two games. AI-assisted work runs under a written process that keeps a person in control.
What we do
- Help your staff use AI tools well. A short session and a one-page standard your team can adopt when you're ready.
- Plan AI work that fits your budget and your data. The right model for each job, chosen for cost and for where your data may go.
- Give your developers gates, not guesswork. A written process for AI-assisted coding, and a starter kit of skills your team keeps.
- Retire the technical debt you've been living with. Old systems rebuilt and documented by an experienced developer, at a cost that finally fits the budget.
- Close the holes before someone else finds them. Vulnerabilities found and fixed, with automated checks so the next one is routine.
NouriLoop: AI built in from the ground up
NouriLoop is a nutrition-tracking web app designed around AI from its first line of code, not one with a chatbot added later. AI powers eight of its features:
- reading a meal, a recipe or a supplement label from a photo
- turning a goal described in plain words into a measurable one, then checking in on it
- explaining glucose responses from a continuous glucose monitor
- "what if" predictions from your own data
- a help assistant that answers from the app's own pages
Underneath, it's built so cost stays a choice:
- Pick the model that fits the budget. Use Anthropic, any of the many models available through OpenRouter, or a model on your own Ollama server with no per-call fees.
- Simple tasks use smaller models. The help assistant runs on a smaller, cheaper model automatically. Only the work that needs a large model gets one.
- Test before trusting a cheaper model. A vision benchmark checks whether a model reads food photos well enough before it's used.
- Every AI call is metered, so the cost of each feature is known, not guessed.
- People review every estimate before it's saved.
We design client systems the same way. No organization should be locked into one AI vendor's prices.
It was built mostly with AI coding agents, with every step planned, directed and reviewed by a person. That's the process our developer course teaches. Open source under AGPL, and live at app.nouriloop.com.
More work
Two browser games, Cranky Coop and Quiet Solitaire, each built with the same review process. Plus open-source library software, in production at Oregon library systems.
Start with a conversation
Tell us what you're trying to do, and we'll tell you whether we can help.