🤖 Lesson 12: Build Your Own AI - Teacher Quick Reference

Print this card for easy reference during the lesson | Evolve AI Institute

📅 Lesson Timeline (90-120 minutes)

10 min
Introduction
15 min
Planning
15 min
Tutorial
20 min
Data Collection
20 min
Training/Testing
20 min
Refinement
15 min
Showcase

🎯 Platform Quick Comparison

🎓 Teachable Machine

Best for: First-time AI creators

URL: teachablemachine.withgoogle.com

Setup: No account needed

Time: Fastest (30-45 min project)

📱 MIT App Inventor

Best for: Mobile app developers

URL: appinventor.mit.edu

Setup: Google account required

Time: Moderate (60-75 min project)

🐱 Scratch with ML

Best for: Interactive games/stories | URL: scratch.mit.edu + machinelearningforkids.co.uk | Setup: Scratch account + ML for Kids account | Time: Longest (75-90 min project)

✅ Pre-Class Checklist

  • Test all platforms beforehand
  • Check camera/mic permissions
  • Ensure stable internet connection
  • Print planning worksheets
  • Set up student accounts (if needed)
  • Prepare backup project ideas
  • Gather physical objects for training
  • Test projection/display system
  • Have troubleshooting guide ready
  • Prepare assessment rubrics

💡 Key Teaching Points

Training Data Quality

  • 50+ samples per category minimum
  • Diversity is more important than quantity
  • Vary angles, lighting, backgrounds
  • Balance samples across categories

Iteration is Normal

  • First attempts rarely perfect
  • 70%+ accuracy is good for students
  • Testing reveals improvement areas
  • Professional AI also iterates

🚨 Common Problems & Quick Fixes

Problem Quick Fix
Camera/Mic not working Check browser permissions in address bar, try different browser (Chrome best)
Low model accuracy (<50%) Add more diverse samples, simplify categories, check for background interference
Student can't think of project Offer menu of pre-approved ideas, show more examples, suggest simple versions
Platform running slow Close other tabs, clear cache, reduce image quality, try different device
Project too ambitious Help identify MVP (Minimum Viable Product), reduce from 5 to 3 categories
Students finish at different rates Have extension challenges ready: improve accuracy, add features, help peers

🎤 Discussion Questions During Class

During Data Collection:

  • "Why do we need so many samples?"
  • "How is your data diverse?"
  • "What makes categories distinguishable?"

During Testing:

  • "What's your accuracy rate?"
  • "Which categories confuse most?"
  • "How will you improve your model?"
⏰ Time Management Tip: Set visible timers for each phase. Build in 5-min buffer between phases for transitions. If running short on time, prioritize: Planning → Data Collection → Training → Testing. Refinement and showcase can be continued next class if needed.

Platform-Specific Quick Guides

🎓 Teachable Machine - 5-Step Process

Step 1: Choose Project Type

teachablemachine.withgoogle.com → Get Started → Choose Image/Audio/Pose

Step 2: Add Classes

Rename "Class 1", "Class 2" to meaningful names → Add more classes (3-5 total)

Step 3: Record Samples

Click Webcam/Microphone → Hold to Record → Capture 50+ samples per class

Step 4: Train Model

Click "Train Model" → Wait 1-3 minutes → Training complete!

Step 5: Test & Export

Use Preview panel to test → Export Model to save or share

Pro Tip: Have students test with a partner before considering their model complete. Fresh eyes catch issues!

📱 MIT App Inventor - Key Steps

Setup (5 min)

appinventor.mit.edu → Sign in with Google → Create New Project

Designer View (10 min)

Add components: Button, Camera, Image, Label → Add AI Extension (Personal Image Classifier)

Blocks View (10 min)

When Button.Click → Camera.TakePicture → Classify → Display Result

Testing (ongoing)

Install MIT AI2 Companion app → Connect → AI Companion → Test on phone

Common Issue: Blocks won't snap together? Check that shapes/colors match. Red triangles indicate errors.

🐱 Scratch + ML for Kids - Two Methods

Method 1: Video Sensing (Simpler)

  • Add "Video Sensing" extension
  • Turn video on
  • Use "video motion" blocks
  • Good for motion-based games

Method 2: ML for Kids (Advanced)

  • Go to machinelearningforkids.co.uk
  • Train custom model
  • Get special Scratch link
  • Import model into Scratch
Time Saver: If time is limited, stick with Video Sensing for first projects. ML for Kids requires two accounts and more setup time.

📊 Assessment Quick Rubric

Criteria Excellent (A) Good (B) Needs Work (C)
Functionality 85%+ accuracy, all features work 70-84% accuracy, minor issues <70% accuracy or major issues
Training Data 75+ diverse samples/class 50-74 samples, some variety <50 samples or not diverse
Iteration 2+ improvement cycles, documented 1 improvement cycle shown No iteration or documentation
Presentation Clear demo + explanation + reflection Demo + basic explanation Demo only, limited explanation

🎯 Differentiation Quick Tips

For Struggling Students

  • Suggest Teachable Machine (easiest)
  • Start with 2-3 categories only
  • Provide step-by-step checklist
  • Pair with peer buddy
  • Offer pre-selected project ideas

For Advanced Students

  • Challenge with 5+ categories
  • Encourage model export/integration
  • Suggest combining AI types
  • Ask to help troubleshoot peers
  • Explore transfer learning concepts
🎉 Celebration Idea: Create a digital gallery of all student projects. Take photos/screenshots during presentations and share with parents/administration to showcase student AI creators!

Need More Help?

Full materials at: /edai/lesson-repository/lesson-12/downloads/

Evolve AI Institute | info@evolveaiinstitute.com