Spam filters
A classifier predicts whether an email belongs in spam by finding patterns in the message.
How classification works ↗Teach your machine how to think.
Hands-on AI literacy for kids aged 9–14
For primary and secondary schools
What is Miltalk?
Miltalk is a handheld AI learning kit that children teach using their own examples. The Vision Kit has a camera for pictures. The Audio Kit has a microphone for sounds. Children choose two categories, collect examples and teach the kit to tell them apart.
Miltalk so far
50children reached
3schools piloted
S$8,000funding received
Why we started Miltalk
Children encounter AI in their apps, classrooms and everyday lives. We started Miltalk to help them understand how a machine learns from data. The concept we teach is called classification: predicting which category something belongs to. Children explore it by teaching a kit with their own examples and checking its answers.
Why teach classification?
It gives children a clear starting point for understanding AI: they can choose the training data, see the predictions and investigate mistakes. The same task appears in familiar tools, from spam filters to image recognition.
Where you see classification
Recognising a category is useful in many familiar AI features. Miltalk lets children explore that idea with a model they train themselves.
Photo labels
Image classifiers can attach labels such as “cat”, “dog” or “flower” to a photo. Google’s ML Kit is one example.
See Google’s image labelling ↗Keyword recognition
A small speech model can classify a short recording as “yes”, “no” or another sound. This is a simple example of recognising spoken commands.
See TensorFlow’s speech example ↗Learn with pictures and sounds
Both kits let children collect examples, train a model and test its predictions.
Miltalk Vision Kit
Train with images
Use its camera to collect visual examples: colour, shape, angle and light all become part of the investigation.
Miltalk Audio Kit
Train with sound
Use its microphone to compare sounds and explore how volume, rhythm and surroundings affect a result.
How it works
Children teach classification by doing it. For example, they label pictures as recyclable or non-recyclable, then show the kit a new object. If its prediction is wrong, they look at the training examples, add better ones and test again.
- 01
Train
Choose your categories
Show the kit labelled examples of two groups. - 02
Test
Try something new
See which category the kit predicts for an unfamiliar example. - 03
Investigate
Find the clue
Compare the wrong prediction with the examples you gave it. - 04
Improve
Change the examples
Add more useful data, then see what happens on the next test.
What kids learn
01
Better examples. Better predictions.
Children see how the examples they choose shape a model’s answers. Narrow training data can lead to mistakes on something new.
02
Put an idea to the test.
Students make a prediction, change one thing and see whether the evidence supports it.
03
Check the answer behind the score.
A score can be high and still be wrong. Students learn to check each prediction against what actually happened.
Early pilot sessions
What we observed
August 2026 · Muhammadiyah Welfare Home
Students kept testing after the activity ended.
During free exploration, one student described the device as comparing a new example with what it had seen before.
October 2026 · Sembawang Secondary School
Students linked mistakes back to the examples.
Students identified incorrect results, proposed better examples and connected the exercise to recognition problems they had observed with school security cameras.
Bring Miltalk to your students
Start with a conversation.
Tell us about your group, preferred dates and what you would like students to explore. We will come back with a practical session proposal.
Enquire about a session