Course Curriculum & Learning Progression
The PictoBlox AI & ML Lab at Axilearn is an advanced, project-based AI programme aligned with NEP 2020. Each level builds on the one before:
- Level 1: pre-trained AI. Learners work with AI that can see faces, bodies, objects, text and markers, and understand speech.
- Level 2: model training. Learners build, test and improve their own models in the PictoBlox ML Environment.
- Level 3: complete AI systems. Learners combine several models with generative AI, image processing and live data into full applications.
Responsible AI Built-In:
Responsible AI (privacy, consent, bias and transparency) is taught in every level.
Entry Requirement:
Basic Scratch or PictoBlox block coding (events, loops, conditions, variables).
Structured Level Breakdown
Level 1
SESSIONS WITH INSTRUCTOR
Perception Pioneers 👁️
Advanced use of PictoBlox's pre-trained AI: face recognition, body and hand landmarks, object detection, OCR, AprilTags, and multilingual speech AI, used to build smart real-world applications.
Prerequisites: Basic block coding
Extensions Used: Face, Body, Object, Vision, OCR, AprilTags (Beta), Speech AI, Translate
Topics Covered:
- How Machines Perceive: rule-based systems vs AI, camera vs stage input, confidence thresholds, bounding boxes, and mapping AI coordinates to the stage
- Face AI: detection, facial landmarks and expression detection, and face recognition by training a face database to identify known people
- Body & Hand AI: pose keypoints and hand landmarks, and calculating distances and angles between points to detect movement
- Object Detection: detecting, counting and tracking everyday objects, and using position and size to make decisions
- Computer Vision: identifying landmarks, brands and objects in images
- Text Recognition (OCR): reading printed and handwritten text from the camera or stage
- Marker-Based Vision: Recognition Cards, QR codes, and AprilTags (tag ID, position, rotation and pose)
- Speech AI: speech recognition, command parsing, text-to-speech and translation, combined into multilingual voice interfaces
- Responsible AI: consent for face data, privacy, and bias in pre-trained models
Mini Projects:
Face-Recognition Attendance System · Air Canvas (fingertip drawing) · Squat & Push-up Counter (pose angles) · Smart Object Counter · Handwritten Math Solver (OCR) · Multilingual Voice Translator · AprilTag Treasure Hunt
Level 1 Project Output:
AI Smart Classroom Assistant. It recognises students by face for attendance, counts raised hands with body detection, reads text from the board with OCR, and speaks announcements in multiple languages.
Level 2
SESSIONS WITH INSTRUCTOR
Model Makers 🧠
Training your own AI. Learners build, test and improve all seven model types in the PictoBlox ML Environment (image, object detection, hand pose, pose, audio, text, and number/regression) and deploy them in block-coded applications.
Prerequisites: Perception Pioneers
Tools Used: PictoBlox ML Environment (All 7 Model Types), NLP, Teachable Machine
Topics Covered:
- How Machine Learning Works: supervised learning, classes and labels, features, and training vs testing data
- Building Good Datasets: webcam capture, uploading and importing datasets, balanced classes, and background or "none" classes
- Image Classifier: multi-class image models and their training settings (epochs, batch size, learning rate)
- Object Detection (ML): labelling bounding boxes and training a custom detector for your own objects
- Hand Pose Classifier: gesture and sign recognition from hand landmarks
- Pose Classifier: full-body pose and activity recognition
- Audio Classifier: recognising sounds and spoken words, and handling background noise
- Text Classifier & NLP: intent detection, sentiment analysis and spam detection
- Number Classifier & Regression: training on tabular CSV data, predicting categories vs predicting values
- Evaluating Models: testing on unseen data, reading training results, and finding and fixing mistakes
- Deploying Models: exporting to block coding, using prediction confidence, and smoothing noisy predictions
- Teachable Machine vs ML Environment: comparing workflows and loading models by link
Mini Projects:
Plant Disease Detector (Image) · Safety Helmet Detector (Object Detection) · Yoga Pose Coach (Pose) · Bird Call Identifier (Audio) · Complaint Categoriser (Text) · Exam Score Predictor (Regression) · Rock-Paper-Scissors AI Opponent (Hand Pose)
Level 2 Project Output:
Sign Language Interpreter. A custom Hand Pose model recognises sign-language gestures, turns them into text and speech, and translates them into other languages in real time.
Level 3
SESSIONS WITH INSTRUCTOR
AI System Architects 🚀
Designing complete AI systems. Learners chain several custom models together, add generative AI with ChatGPT, pre-process images with the Image Processing extension, feed live data into predictions, build multiplayer AI experiences, and deliver a competition-ready capstone.
Prerequisites: Model Makers
Tools Used: ML Env, ChatGPT, Image Processing, Weather, Data Logger, Cloud Var, IFTTT
Topics Covered:
- Multi-Model Pipelines: chaining models and extensions, decision logic, and combining the confidence of several models
- Generative AI with ChatGPT: prompt engineering, role and context prompts, and building conversational agents with Speech Recognition and Text to Speech
- Limits of Generative AI: hallucinations, fact-checking and safe prompting
- Image Processing: blurring, thresholding, and contour, circle and polygon detection, used to clean up images before ML
- Spatial AI: AprilTag pose tracking for AR-style and robotics-style applications
- Data-Driven AI: feeding live weather data into regression models, logging predictions with Data Logger, and live charts with the Graph extension
- Connected AI: multiplayer AI games with Cloud Variables, and AI-triggered alerts with IFTTT Webhooks
- Improving Models: error analysis, collecting hard examples, class imbalance, overfitting and retraining
- Real-Time Performance: confidence tuning, reducing false positives, and running several models at once
- Responsible AI Design: bias audits of your own models, privacy by design, explainability, and model and dataset cards
- Capstone Studio: defining the problem, mapping it to the UN Sustainable Development Goals, prototyping, user testing and pitching
Mini Projects:
AI Fitness Coach (pose + ChatGPT feedback) · Shape & Coin Sorter (Image Processing + ML) · Smart Crop Advisor (weather data + regression + ChatGPT) · Multiplayer Gesture Battle (Cloud Variables) · Intruder Alert System (face recognition + IFTTT) · Voice-Powered AI Tutor
Level 3 Project Output:
AI for Social Good Capstone, an original AI system that uses at least two custom-trained models, at least two AI extensions and live data logging. Learners submit it with a dataset card, an accuracy report and a demo video, ready for AI competitions such as Codeavour.
Course Summary
Target Age:
Ages 11–17
Levels:
3 Levels: Perception Pioneers → Model Makers → System Architects
Total Duration:
34 Weeks (~68 hours)
Format:
Live + Lab
Coding Mode:
100% Block Coding
Platform:
PictoBlox Desktop App
Requirements:
Laptop/Desktop, Webcam, Mic & Internet
Certificate:
Axilearn Certified (Each Level)
Key Highlights:
- A fully AI- and ML-focused curriculum with no basic coding filler
- 12+ AI extensions: Face, Body, Object, OCR, Speech & ChatGPT
- Train all 7 ML model types: Image, Object, Pose, Audio, Text, Regression
- Multi-model AI systems with generative AI & live data
- Model evaluation, error analysis & retraining
- Responsible AI: privacy, bias audits & dataset cards
- 20 advanced mini projects & 3 level-end projects (Codeavour capstone)
Lab Requirements:
Cloud extensions (ChatGPT, OCR, Speech) require an active internet connection. Image Processing operates in Stage Mode. Parental consent required for face recognition data.
