Training AI on Appalachia
Actual General Intelligence
First-person footage of skilled tradespeople doing real work. The knowledge that isn't in the training data.
The Problem
Last summer I built a fort with ChatGPT. 200 hours of conversation. Framing, fasteners, layout — it told me everything.
Never mentioned OSB has a moisture vector. Sheathing's engineered for buildings with HVAC. My fort has no AC. It's delaminating.
Any carpenter would've said it in five seconds:
"That's not a house. Different rules."
Training data is manuals and specs. It's not the guy who's watched weird shit fail for 30 years.
Blue Collar Raw — The Product
Hard hat cameras on tradespeople while they work and teach. First-person perspective of hands doing the work.
Whisper transcription. The narration IS the metadata. Self-labeling data from natural teaching.
Faces and locations stripped. Tagged by task, tool, and duration. Ready for ingest.
Not lab environments. Not college kids folding laundry. Real tradespeople solving real problems.
The Depth
Placeholder footage — Blue Collar Raw
The Buyers
A robot learning to wire a house needs to see a wire nut installed a thousand times. Not from a manual. From a guy who's done it ten thousand times and can feel when it's wrong.
— The pitchSilicon Valley trained AI on the internet.
We're training it on Appalachia.
Sample dataset available January 2026.
10 trades. 10 days. 3000+ hours.
Contact us for your requirements.