It learns the way a curious child might: by seeing a shape, hearing a sound, and noticing they belong together.
This is a tiny computer program — not a person, not magic — that is learning the alphabet one letter at a time. You can watch it learn, and you can teach it.
Show it the letter A and it will say the sound and draw the shape back to you. Play it a sound and it will draw the letter it heard. Draw your own letter with your finger or mouse and watch it try to guess — and sometimes get it wrong, which is a completely normal part of learning.
It runs entirely in your web browser. There is nothing to install, no account to make, and nothing about you is collected. It is free and it is yours to explore.
Press A. The program looks at the shape, then says the sound and draws the letter it recognises.
Press a sound button. It listens, then makes a picture of the letter it heard — from memory.
Draw a letter yourself and press Test it. Watch it guess. When it misses, ask your child why — that is the fun part.
Every letter is taught in three ways at once — the Latin shape A, its English sound, and the Hindi (Devanagari) shape अ.
It doesn't keep them in separate boxes. It notices that a shape and a sound keep appearing together, and links them — the same simple idea a child uses when they learn that the squiggle “A” and the sound “ah” go hand in hand. Because it learns them side by side, if you play it a sound it might draw you the English letter or the Hindi one — it knows both.
Each page is a different way to look at the same little program. Start anywhere.
Press a letter, hear its sound, watch it draw. Or draw your own letter and see if the program can guess it. Best for kids and first-time visitors.
A small program with its own eyes, ears, hand, and voice. It teaches itself the alphabet in a few seconds, then you can show it letters, teach it your own voice, and even give it a bigger eye to see with. Nothing about you is uploaded.
Draw any letter and three eyes (tiny, medium, big) each look at it, remember it, and draw back what they remember. A hands-on way to see why eye size matters for how well you can see.
For the curious: a plain-English tour of how a shape and a sound become the same kind of "signal" inside the program, and exactly what it can and cannot do.
No rush, no scores, no “levels”. Just a quiet little machine you and a child can poke at together and talk about.
Getting it wrong is a doorway, not a dead end. Every miss is a chance to notice how careful reading really is.
This program learns by the same kind of process a child's brain is thought to use: it makes tiny predictions, notices when they miss, and updates. That idea — brains as prediction machines, learning by minimising surprise — is called active inference, and it is one of the leading scientific accounts of how living things come to understand the world. The history of this idea goes back to Ibn al-Haytham and Helmholtz.
What UNI is not: a body, a life, feelings, or a story of its own — so it does not mean a letter the way you do. What today's mainstream "AI" — ChatGPT and other Large Language Models (LLMs) — actually is: a very large statistical machine that predicts the next token in human writing. It does not perceive a world. It does not act in one. It does not ground words in anything that could be right or wrong about reality. It burns enormous amounts of energy to sound convincing — sounding like understanding is not the same as understanding. UNI is built the opposite way: observe nature, measure the gap between prediction and observation, reduce the gap, and report durable, reproducible, falsifiable evidence for every claim. Natural intelligence is less costly to the universe than pretending to play God with gigawatts.
We try to be careful, not clever. This is a learning demonstration, built on the same mathematics that a long line of scientists — from Helmholtz to Friston — have proposed for how brains actually learn: predicting what will happen, noticing when they miss, updating. It uses transparent counting instead of the giant neural networks behind today's LLMs — because you can check every step, and because natural intelligence does not need gigawatts to grow a mind.
What today's mainstream "AI" is: ChatGPT and other LLMs — enormous next-token predictors trained on human writing. They eat tokens, burn massive power, and do not ground words in a world: sounding like understanding is not the same as understanding. What UNI is: a small, transparent implementation of active inference — observing nature, measuring the gap between prediction and observation, reducing the gap, and reporting durable, reproducible, falsifiable evidence for every claim. What it is not: a child, an embodied life, or a first-person point of view — it does not mean anything by "A". But every number we publish about it is one you can rederive on your own laptop.
If you're the technical sort, there's more to see. The Observatory shows how a shape and a sound become the same kind of "signal" inside the program. The Being page shows the same little program doing much more (teaching itself, hearing your voice, drawing with different-sized eyes). The Eye Lab lets you compare a tiny eye and a big eye side by side. And the whole thing is built to be pressure-tested — you can try to break it and see where its honest limits are.