BlueberryML · London

Manifolds, not prompts.

BlueberryML builds the geometry AI thinks in. We learn the shape of human domains — how people learn, how people feel, who a character is — and license it, so AI systems reason over structure instead of improvising from a prompt.

See how it works →Work with us

The learning manifold is in production in schools across India. →

How it works

Knows where it is. Decides fast. Talks well.

A language model on its own is fluent but has no sense of place: every turn starts again from a prompt.

A manifold on its own knows exactly where someone is, but can't say a word.

Put a fast System 1-type decision layer between them, and you get a system that knows where it is, decides quickly and talks well. Neither half can do that alone.

How the method works →

The manifold

slow · learned · persistent

Where are we, and where should we go next?

System 1

fast · calibrated · typed

What happens now? On every turn, in milliseconds.

The language model

fluent · flexible · steered

How do we say it?

Every turn is evidence. Evidence moves the position.

The manifolds

One geometry of a whole life.

How we develop as children, who we are, how we change as we age. One method, applied domain by domain.

In production

In production

At national scale, today.

The learning manifold is in production in India, where our client Chrysalis uses it across its schools.

What's in production →

~600,000

children in Chrysalis schools

Sovereign by design

We never touch your data.

We help you train a manifold where your data already lives, then license it to you to run there. The data always stays sovereign: under your control, in your jurisdiction, and yours.

Trained where your data lives.

We help you train your manifold on your own infrastructure, in your own jurisdiction. The data never leaves your control, and it never comes to us.

Then licensed to you.

Once it’s trained, you run the manifold locally under licence, under your brand. Learner records, voices, transcripts and results stay with you, and you own them.

The models around it, too.

We help build the rest of the stack, such as speech recognition, voice and scoring models, to run on your own hardware. That’s what makes sovereign AI possible.

How we handle data →

Built on the manifold