
CR Innovations Lab
Corey Ryan's lab notes on local AI, fine-tuning, and small-business automation
I'm Corey Ryan. I run Ryan² Design Labs, where we build AI employees for small businesses: voice agents that answer the phone, follow-up that never forgets, and the plumbing that connects them to the tools a business already uses.
This domain is the lab side. Everything here runs on hardware I own: consumer GPUs, a 12-year-old server, and a lot of measurement. I write up what I learn, including the parts that didn't work.
How I work
Self-taught, decades in. I'd rather measure than guess, and I'd rather build the thing than read another post about building the thing. Rules get a fair hearing and then I try it anyway. If I tell you something works, there's a number behind it, and if I was wrong, that gets written up too. Salt Lake City, with mountains close enough to think in.
What's on the bench
- The article series — engineering write-ups on local inference, training-data curriculum, and the mistakes that cost real GPU hours. Seven so far. Read them.
- WILLAI — a self-improving agent harness: every task is decomposed into testable criteria, proven with evidence, and the misses are captured as training data. It runs my day.
- Local inference stack — an orchestrator that routes requests across a few consumer GPUs, with a 0.5B gate model deciding when the 14B is actually needed. The benchmark write-ups come from this.
- AI Studio — a browser workspace for planning, drafting, and shipping content with local and hosted models side by side.
Latest writing
-
I Split My Training Data Into 5 Layers — Here's Why Order Matters
How curriculum learning from a 2009 paper changed the way I train a 4B model on a single GPU
-
I Gave My AI a Brain — Here's How It Learns From Every Conversation
How I built a self-improving memory system from pgvector, markdown staging files, and a feedback loop that retrains my local model
Elsewhere
- LinkedIn — where the articles get discussed
- Ryan² Design Labs — the business
- Omnyka — co-founder and CTO. AI help desk agents for MSPs, built on locally run, custom-trained models.


