Skip to main content

Stage 10: Present an Honest AI Demo

Course progressStage 10 of 10
~90 min
Your workspace

Keep your Colab notebook tab open all session. Open in a new tab — don’t use the buttons in this page to leave the course.

Train your AIGoogle ColabOpen in a new tab ↗No-code warm-upMachine Learning for KidsOpen in a new tab ↗
Build

a final results table and a two-minute demo script

Learn

how to present model behavior honestly, including limits and mistakes

Ship

a showcase-ready notebook with evidence, not just lucky examples

Teacher demo

Model a short demo: one success, one mistake, one honest limitation. Keep it under two minutes.

The big idea​

A strong AI demo does not pretend the model is perfect. It shows evidence: test accuracy, examples, confidence, wrong predictions, and a limitation.

Today you turn the notebook into a story another person can understand.

Your final demo should follow a simple evidence rubric: state the claim, show the test score, show a correct case, show a wrong or uncertain case, name a limitation, and suggest a next experiment.

How the Python ML workflow connects
  1. 1
    Photos / CIFAR-10labeled image examplesStage 1
  2. 2
    Notebook variablesx_train, y_train, class_namesSetup-2
  3. 3
    Prepared datanormalized pixels and fair pilesStage 3
  4. 4
    Keras modelCNN layers and summaryStage 4
  5. 5
    Training historyepochs, loss, accuracyStage 5
  6. 6
    Test evidencesealed score and mistakesStages 6-7
  7. 7
    Improved modelaugmentation comparisonStage 8
  8. 8
    Inferenceuploaded image to top-3 guessesStage 9
  9. 9
    Demo evidencetable, confidence, limitationStage 10

Stage 10 turns the whole workflow into a story. The final demo should connect data, model, training, test evidence, predictions, and limitations.

New words
evidence
results that support a claim
limitation
something the model cannot do well yet
demo script
a short planned explanation of what you built
Before you start

You need a working final_model, load_and_preprocess_image, and at least 3 uploaded images from Stage 9.

Build it​

Step 1 — Choose demo images​

Pick 3-5 images:

  • at least one clear success
  • at least one tricky or wrong example
  • at least one image you can explain using confidence or top-3 guesses

Step 2 — Build a results table​

demo_images = [
    {"path": "airplane.jpg", "actual": "airplane"},
    {"path": "truck.jpg", "actual": "truck"},
    {"path": "my_cat.jpg", "actual": "cat"},
]

results = []

for item in demo_images:
    image_array = load_and_preprocess_image(item["path"])
    scores = final_model.predict(image_array)[0]
    predicted_index = scores.argmax()
    predicted_name = class_names[predicted_index]
    confidence = scores[predicted_index]

    results.append({
        "image": item["path"],
        "actual": item["actual"],
        "prediction": predicted_name,
        "confidence": confidence,
        "correct": predicted_name == item["actual"],
    })

results

This table is your demo evidence.

Step 3 — Display each image with its result​

for result in results:
    plt.figure(figsize=(3, 3))
    plt.imshow(plt.imread(result["image"]))
    status = "correct" if result["correct"] else "wrong"
    plt.title(f"{result['prediction']} ({result['confidence']:.0%}) - {status}")
    plt.axis('off')
    plt.show()

Include at least one wrong or uncertain result. Honest demos are stronger.

Step 4 — Write your final explanation​

In a text cell:

My AI demo script:
1. I trained an image classifier on CIFAR-10, a dataset of 10 image classes.
2. My honest test accuracy was about _____%.
3. My model is strongest at __________.
4. It struggles with __________, as shown by __________.
5. One limitation is that it can only choose from the 10 classes it learned.
6. One improvement I would try next is __________.

Step 5 — Practice the two-minute walkthrough​

Read the script while showing your notebook evidence:

  1. dataset shape or image examples
  2. training chart
  3. test accuracy
  4. wrong prediction or hardest class
  5. final demo table

Understand it​

The goal is not to make AI look magical. The goal is to explain what it learned, how you measured it, and where it fails. That is real machine-learning communication.

A lucky gallery can hide problems. An evidence table shows what happened and lets your audience trust you.

Try this​

Learning beat

Try this

Three short experiments. Predict before you run, then test your guess.

Predict first

Before showing a parent or peer, predict which result they will ask about first. Is it a success, a mistake, or the accuracy score?

Compare

Compare a demo with only correct images to a demo with one wrong image and an explanation. Which one sounds more honest?

Connect

Which stage taught the most important idea for your final explanation: data, model, training, evaluation, errors, augmentation, or inference?

Test your stage​

  • Your results table includes image, actual label, prediction, confidence, and correct/incorrect.
  • Your demo includes at least one wrong or uncertain example.
  • Your script includes test accuracy and one limitation.
  • Workflow check. Point to the workflow map and explain the full path from CIFAR-10 data to demo evidence.
  • Evidence check. A peer or coach can point to the score, the correct case, the mistake, and the limitation in your notebook.
  • Design check. You can explain your model in two minutes without saying "it just knows."

If it breaks​

  • results is empty. Check demo_images has valid paths.
  • Every image is wrong. Use clearer examples for the demo, but keep one mistake if possible.
  • A path disappears. Colab uploads vanish after disconnects; upload again.
Coach notes

Require the limitation. That is what turns the final project from a flashy notebook into an actual course outcome.