How it works · Saraswati

A teaching engine of our own. Open, and always learning.

Saraswati is Zavmo’s own small language model, named after the goddess of knowledge. It doesn’t rent intelligence from a giant black box. It’s built on open foundations, trained only on genuinely good teaching, and designed to get a little better the more people it helps.

Built on open foundations Never trained on your words Held to the neurogogy bar

A teaching engine of our own. A small, focused voice that reaches every learner.

Small, open, ours

Most AI learning rents a giant black box. We built a small, open teacher instead.

Bigger doesn’t mean better at teaching. Focused does. Saraswati is built on open foundations we can see inside and control, not a frontier model we’d have to rent and can’t. That means teaching you can actually trust, costs low enough to put learning within reach, and an engine that’s genuinely ours, with no vendor able to change the rules or pull the rug.

  • Open foundation
  • Runs lean
  • No lock-in
  • Ours, not rented
See the tech

Saraswati is built on Phi-3 Mini (3.8B parameters, MIT-licensed and openly available) with small, swappable per-market adapters (LoRA). It’s cheap to run and carries no dependency on a frontier-model vendor for the teaching itself. That’s the reason we can keep it accurate, controllable and affordable.

How it’s trained

It doesn’t learn from the internet. It learns from great teaching.

A chatbot is trained on the whole web and hopes for the best. Saraswati is trained only on teaching that earns its place, measured against the science of how people actually learn.

01

Held to the Neurogogy bar

Every lesson Saraswati learns from has to pass a strict quality test built from Neurogogy’s pillars, dozens of checks for whether it’s actually good teaching. Fall short, and it never makes the cut.

What is Neurogogy? →
02

Steered by what really helped

It learns to do more of what genuinely moved real learners: real progress, staying in flow, coming back, getting curious. It keeps the struggles too, because that’s where the most useful lessons hide.

03

Never trained on your words

The line it won’t cross: Saraswati never learns from what you actually type. Only the shape of what helped, never your text. Your words stay yours, full stop.

04

Tuned for where you are

It’s fine-tuned market by market. The UK’s way of recognising skills isn’t Singapore’s or Australia’s. So the teaching fits your world, not some global average.

Every improvement has to earn its way in

Saraswati distils what worked into a better version of itself, then has to prove it teaches better on a fresh test before it’s allowed anywhere near a real lesson. If a new version ever slips, it rolls back on its own. Getting better is the entire point, but never at your expense.

Rolled out carefully. We switch these improvements on slowly and behind safety gates. Better is earned, not rushed, and the person in front of it always comes first.

Saraswati is Neurogogy, put to work and kept sharp

Neurogogy is the science of how the brain learns. Saraswati is the engine that teaches by it, and gets better at it over time.

Read about Neurogogy →

Ready when you are

Teaching That’s Ours, Open, and Always Improving

See the route to any UK role, qualification or skill, then walk it with your companion.

Browse the catalogue