The model
Powerful — and frozen. No learning, no memory, confidently wrong.
// post-training enhancements
We make frontier AI
work in a real factory.
The strongest models in the world are frozen at training — no learning, no memory, no grounding in the operation, no way to digest industrial-scale live data within real limits of context, latency and cost.
tinyAI builds the post-training enhancement layer that closes those gaps — and runs it live in a real factory.
95%
of enterprise AI pilots return nothing.
Not because the models aren’t impressive — because a raw model can’t survive contact with a real operation. That gap is exactly what we own.
// roughly — MIT Project NANDA, 2025
// the triangle
One problem, three corners.
The data & knowledge
A flood of live data — and the human knowledge that gives it meaning.
The user
A real worker, mid-shift, expecting a deep, correct answer in seconds.
The product lives in the middle of this triangle.
// the enhancement layer
What we add to a raw model.
Memory
It knows the operation — and keeps learning it.
Truth
Every answer anchored to measured data, sources included.
Hands
Connected to the machines, the sensors, the ERP. It looks, not guesses.
Scale
Factory-scale data, deep answers in seconds — at sane cost.
Initiative
It investigates around the clock, and speaks up first.
Presence
On WhatsApp, in Hebrew, where the workers already are.
// what drives us
We live this triangle every day.
tinyAI runs live inside a real factory — real machines, real workers, real questions, every day. What that daily contact teaches can’t be learned from titles or demos. It’s the experience we compound into the product.
// the trajectory
From answering to acting.
Every level is earned the same way — grounded, sourced, trusted.
- 01
Answers
Any question, answered straight from live operational data.
- 02
Recommends
Owns the recurring calls — recommendations denominated in money.
- 03
Automates
Handles the repetitive decisions and actions.
- 04
Autonomy
When allowed, it takes control of parts of the line itself.
Live today
Running on a real production line — right now.
In production on real machine and ERP data, answering in Hebrew on WhatsApp every day. The same layer travels to any operations-heavy industry.
Field Notes
How we think in the open.
Why We Build One End-to-End System, Not a Piece
Most factory AI dies in the pilot — not because the technology fails, but because someone ships a piece and leaves a plant with no AI team to make it all work together. We build the whole thing, end to end, and run it.
The Semantic Layer Is the Product, Not the Plumbing
Everyone in data sells a 'semantic layer' now. In a factory it means something different — and harder: a model of what the data actually means, which is the difference between an answer that is right and one that is only confident.
How We Deliver Story to a Factory Agent
Retrieval is useful, but operational AI systems need more than a bag of context. In Factory Agent, we separate stable identity, tribal knowledge, bounded snapshots, conversational contracts, and living story objects so the model receives the right shape of truth for the job.
View all posts →Contact
Working on the same problem?
If you’re building, researching, or deploying AI against the real world — we’re happy to compare notes. And if you run an operation, we’ll talk too.
journey@tinyai.studio Based in Israel. Working from the floor outward.