OMO AI: learning games that teach kids ideas before numbers

Children learn to fear learning early.

OMO AI is building an agentic AI that turns imagination into the gateway to abstraction: immersive, gamified worlds for foundational literacy and numeracy (FLN) in Class 2–4. Kids learn the ideas (quantity, pattern, sequence, language, reasoning) before they meet letters and numbers.

Built for India's first-generation and multilingual learners. Inclusive design with neurodivergent learners in mind is next.

ASER shows whether children can read and calculate. OMO targets the deeper gap: whether they understand the ideas well enough to use them.

Gaps open in the first years of school and widen from there. When learning arrives as rote, pressure and punishment, many children decide they are simply "stupid" at maths, long before they meet the ideas behind it. OMO exists so that no child learns to hate learning. We start with play, where being wrong is just the next try.

The ideas OMO teaches, without numbers.

Numeracy

• LiveChance"This bag will probably give me red." Pick the Bag
• In designQuantity"I can see there are three without counting."
• In designComparison"This plate has more, I can tell by pairing them up."
• PlannedFair share"Everyone got the same, so it's fair."
• PlannedPart–whole"These two pieces make the whole roti."
• PlannedPattern"I know what comes next."

Literacy

• PlannedSounds"These two words start the same."
• PlannedSequence"First this, then that, then the end."
• PlannedStory"I can tell you what happened and why."
• PlannedLanguage"I know this word in Hindi and in English."

How it works, built with Claude.

Today

Claude designed and generated every level of Pick the Bag, in English and Hindi, and helped build this site and the game engine. In the game, difficulty adapts with simple rules in the browser: the game reads play signals (choices, hesitation and retries; rule-based today, Claude-powered with API credits), and the next level gets harder when the idea clicks and gentler when it hasn't.

1. ConceptAn idea a child should feel, e.g. "more likely".
2. →
3. ClaudeWrites levels and Hindi/English hints from a short brief.
4. →
5. Level JSONBags, target colour, answer, concept, hints.
6. →
7. ValidationA script checks the right bag is truly most likely, with no ties.
8. →
9. Kid testing5–10 children play each build at home, parents present.

{
  "id": 13,
  "tier": 2,
  "concept": "Proportion beats count: more of a colour is not enough, the others matter too",
  "target": "red",
  "bags": [
    {
      "objects": [
        {
          "colour": "red",
          "count": 3
        },
        {
          "colour": "blue",
          "count": 1
        }
      ]
    },
    {
      "objects": [
        {
          "colour": "red",
          "count": 5
        },
        {
          "colour": "blue",
          "count": 5
        }
      ]
    }
  ],
  "correct_bag": 0,
  "chance": [
    "3/4",
    "1/2"
  ],
  "hint": {
    "en": "More red is not enough. Look at the other balls too!",
    "hi": "सिर्फ़ ज़्यादा लाल काफ़ी नहीं। बाकी गोले भी देखो!"
  }
}

Next Planned · needs API credits

1. Play signalsChoices, hesitation, retries.
2. →
3. Learner stateWhat the child already understands.
4. →
5. ClaudeSelects or creates the next level for this child.
6. →
7. Adult noteA plain-language note for parents and teachers.

Responsible AI for children.

OMO is built for 6–10-year-olds, so safety is part of the design, not an add-on.

• Children never chat with the AI. Claude works behind the scenes: it generates structured game levels (bags, colours, hints), not open conversation.
• Everything is validated before a child sees it. A script checks every generated level (the right answer is truly correct, hints are short and age-appropriate), and we review content before it ships.
• Adults are present. Play is supervised by a parent or teacher, at home or in class.
• No personal data from children. We don't ask children for names, photos, locations or contact details. On this website, play stays in the browser and nothing is sent anywhere.
• AI is disclosed to adults. Parents and teachers are told clearly where and how AI is used in OMO.
• Parents are in control. Parents can ask to see or delete any data at any time: satyam@omoai.lol.

We design OMO in line with Anthropic's Usage Policy for products used by minors, and we'll keep strengthening these safeguards (including for future features like the companion character) before any wider launch.

Tested by kids.

Daksh is 8, in Class 3, and has never been taught probability. In our game he picked the bag most likely to win, by feeling "more likely" before knowing any symbol.

Every build is tested by 5–10 children at home, with parents present.

What we've seen

• Kids replay without being asked.
• They grasp "more likely" before they meet the symbol.
• Hindi and mixed-language hints hold their attention better.
• Pace varies hugely at the same age, which is why we're making it adaptive.

Where OMO is going. Concept

चलो, थैला चुनो!
Let's pick a bag!

Progress note

Daksh understands more likely.

Next: comparing chances.

How OMO will reach children. Plans

Safety and questions.