3-Day Artificial Intelligence Workshop
Analyse • Evaluate • Engineer • Build • Test • Defend
The Class 10 AI Adventure Workbook takes the Class 9 curriculum one level deeper. Students move beyond describing AI and focus on analysing data, evaluating models, debugging code, identifying failure modes, designing systems and defending technical decisions. The three-day journey is organised as Analyze & Code → Explore & Create → Build & Innovate.
Implementation Note: The learning topics, activities and project concepts below come directly from the Class 10 workbook. Where the workbook does not prescribe a particular software platform, suitable recommended implementation tools have been added for practical workshop execution.
Day 1 — Analyze & Code
AI Foundations, Data Auditing & Python
Day 1 develops the technical foundation needed to judge and build AI systems rather than simply use them.
Students Learn
- Rule-based vs data-driven systems
- Human vs machine intelligence
- Input → Process → Output → Feedback
- AI vs Machine Learning vs Deep Learning
- Algorithms and decomposition
- Algorithm efficiency
- Structured vs unstructured data
- Features and labels
- Useful, unfair and leaking features
- Data leakage
- Data cleaning
- Missing values and duplicates
- Data quality
- Sampling, historical, label and measurement bias
- Dataset design and balance
- Python variables and data types
- Input validation
- Lists, indexing and slicing
- Decisions and loops
- Functions and return values
- Strings and text processing
- Debugging
The workbook contains seven Python labs built around Predict → Debug → Modify, with students expected to understand and repair code instead of simply copying working examples.
Day 1 Activities
AI System Detective • Who Should Handle This? • Smart System Design • Algorithm Architect • Feature Engineer • Data Detective • Bias Detective • Dataset Design • Python Debugging Challenges • Marks Analyzer
Students also work with deliberately messy data containing impossible values, duplicate rows, missing data, wrong units and inconsistent coding, learning how each problem can affect an AI result.
Day 2 — Explore & Create
Python, Machine Learning Evaluation & Generative AI
Class 10 students don’t just ask whether a model works. They learn to measure how well it works and decide whether it should actually be used.
Students Learn
- Python analysis on real datasets
- Searching and pattern detection
- Student Performance Analysis
- The eight-stage Machine Learning workflow
- Training, validation and test data
- Classification, regression and clustering
- Accuracy and error
- Model evaluation
- Overfitting and underfitting
- Error distribution
- Model comparison
- Generative AI
- AI-generated text, images, music and video
- Hallucinations
- Expert Prompt Engineering
- Responsible AI and accountability
- Verification of AI-generated claims
Students calculate accuracy from prediction results, compare models and defend which model they would actually use.
Class 10 Prompt Framework
Students move from four-part prompting to a five-part framework:
Role + Context + Task + Format + Constraints
Constraints make the result testable—for example, requiring sources, marking unverified information, showing calculations or limiting the response to supplied information.
Day 2 Activities
Python Data Analysis • Student Performance Analyzer • Eight-Stage ML Workflow • Model Evaluation • Model Judge • Prompt Arena • Responsible AI Scenarios • Level 3 AI Quiz
The Prompt Arena includes four rounds: Explain, Solve, Critique and Audit. Importantly, students write evaluation criteria before seeing the AI output and receive points for finding and proving incorrect AI claims.
Day 3 — Build & Innovate
AI Engineers & Innovators
Day 3 moves from evaluation to engineering.
Students must specify, build, test and defend an AI system rather than simply present an idea.
Students Explore
- Computer Vision
- Image classification
- Classification vs object detection
- Confidence thresholds
- Computer Vision failure modes
- Natural Language Processing
- Tokenisation and intent recognition
- Sentiment analysis
- Chatbot design
- Robotics
- Internet of Things
- Sensors and actuators
- Safety and fail-safe design
- System specification
- Input → Process → Output
- Hostile testing
- Failure-mode analysis
- Prototype development
- User feedback
- Live demonstration
- Innovator Pitch & technical Q&A
The final challenge requires a written specification, a working component, hostile testing and an explanation of the system’s limitations.
Projects Students Build & Explore
1. Quiz / Marks Analyzer
Students combine their seven Python labs into one program built from a requirements list.
The program is expected to handle real user behaviour and survive tests such as:
- Empty data
- Impossible marks
- Negative values
- Unknown student names
- Lowercase names
- Identical marks
Students use functions for tasks such as summarising data, assigning grades, searching and generating data-based advice.
Implementation Tools:
Python
Recommended Platform:
Replit / Google Colab
2. Student Performance Analyzer
Students develop a more advanced Python program using three test scores per student.
The system must include:
- Five or more functions
- Student averages
- Best test
- Performance trends
- Class average and spread
- Number of students improving
- Search by student name
- Flags for students slipping or below 40
- Mark validation
- Handling of missing scores
- A stated limitation of the report
Implementation Tools:
Python
Recommended Platform:
Replit / Google Colab
3. Computer Vision — Specify a Classifier
Students design a proper image-classification system specification.
They decide:
- 3–5 mutually exclusive classes
- How many training images are required
- Where the data comes from
- How varied the images should be
- The confidence threshold
- What happens below that threshold
- The most likely failure mode
- How failures could be detected after deployment
They also explore problems such as distribution shift, unseen classes, shortcut learning and uneven accuracy.
Recommended Implementation Tool:
Google Teachable Machine
Additional Tools:
Phone camera / webcam for collecting and testing images
This lets students turn the workbook’s classifier specification into a working no-code machine-learning demonstration.
4. Study Assistant — With Refusals
Students design a focused study assistant that handles a small number of tasks well rather than pretending to answer everything.
Students specify:
- A particular user
- Four useful intents
- Keywords and possible phrasing
- Responses
- Where each intent may misfire
- An
UNKNOWNpath - Two hard refusals
- A disclosure that the user is speaking to a bot
- Situations where the assistant should admit it does not know
The workbook specifically asks students to design refusals and unknown-intent behaviour because these are important parts of real chatbot safety.
Implementation Tools:
Python – for the workbook’s intent-matching logic
Recommended Tools for a richer prototype:
Botpress – structured chatbot development
ChatGPT / Gemini – conversational AI functionality
5. Smart Device — With Failure States
Students design a smart device around:
Sense → Process → Act
Class 10 adds the important question: What happens when the system is wrong or a sensor fails?
Students define:
- Sensors
- Processing logic
- Actuators
- Real decision thresholds
- A fail-safe state
- What the device must never do
- What happens if a sensor fails
- The safe default
- How a human can stop it
- Who must be informed that the device exists
The workbook example is a Lab Door Occupancy Monitor using door, motion and CO₂ data.
Recommended Implementation Tools:
micro:bit • MakeCode
Optional Hardware:
Motion sensor • Temperature sensor • Light sensor • Distance sensor • LEDs • Buzzer
Computer Vision Exploration
Students learn the Class 10 computer-vision pipeline:
Capture → Preprocess → Features → Classify → Decide
They learn that engineering does not stop when a model returns a prediction—the system must also decide when the confidence is high enough to act and when a human should be asked instead.
Recommended Implementation Tools:
Google Teachable Machine – classifier development
Phone / Webcam – image capture and testing
NLP & Chatbot Exploration
Students study the language-processing pipeline:
Text → Normalise → Tokenise → Intent → Response
They learn why simple keyword systems fail because of:
- Negation
- Word order
- New or unexpected phrasing
- Sarcasm
- Mixed sentiment
They also build a Python intent matcher that can return UNKNOWN when confidence is insufficient.
Implementation Tools:
Python – intent matching and text processing
Recommended: Botpress / ChatGPT / Gemini for conversational prototypes
Robotics & IoT Exploration
Students investigate how AI interacts with the physical world through:
Sense → Process → Act
They also study four important safety principles:
Fail Safe • Physical Stop • Bound the Actuator • Log Every Decision
The workbook emphasises that robotics requires planning for what happens when sensors, software or networks fail.
Recommended Implementation Tools:
micro:bit • MakeCode
Generative AI Creation
The workbook explores four forms of Generative AI.
AI Text
Explanations, summaries, drafts, code and test questions.
Recommended Tools:
ChatGPT • Claude • Gemini
AI Images
Illustrations, posters and concept art generated from descriptions.
Recommended Tools:
DALL·E • Canva AI
AI Music
Backing tracks created from descriptions of style, tempo and instrumentation.
Recommended Tool:
Suno
AI Video
Short clips created from written scene descriptions.
Recommended Tool:
Runway
The workbook stresses that Generative AI produces plausible continuations rather than verified facts, which is why fluent output still needs checking.
Final Project — AI Innovation Challenge
The main Class 10 challenge follows:
Specification → Build → Test → Defend
Unlike the earlier classes, students must create a written specification before writing code and identify their system’s failure modes themselves.
The 12-Step Process
1. Find the Problem
Choose something real, recurring and specific.
2. Define the User
Identify and interview a named person.
3. Write the Problem Statement
Include a measurable outcome.
4. Write the Specification
Define input, output, success criteria and limits.
5. Design the System
Plan Input → Process → Output.
6. Choose the Technology
Select one appropriate technology and justify it.
7. Build the Component
Create something rough, working and demoable.
8. Test Hostilely
Try deliberately bad inputs.
9. Get Feedback
Test on someone outside the team.
10. Name the Failure Modes
Identify at least two ways the system can break and who will be affected first.
11. Revise
Improve the worst problem discovered.
12. Defend It
Run the live demo and answer technical questions.
Hostile Testing
Class 10 introduces a particularly practical project requirement: students deliberately try to break their own system.
Tests include:
Nothing entered • Out-of-range input • Wrong data type • Unknown class • Ambiguous input • Same input repeated
Students record whether the system crashes, produces a wrong answer or handles the situation correctly.
Suggested Implementation Tools for the Innovation Challenge
Students select technology based on the actual problem rather than forcing AI into the solution.
Python / Data System: Replit or Google Colab
No-Code ML / Classifier: Google Teachable Machine
NLP / Chatbot: Python, Botpress, ChatGPT or Gemini
Smart Device: micro:bit + MakeCode
Generative AI: ChatGPT / Claude / Gemini
AI Images: DALL·E / Canva AI
Prototype Interface: Figma / Canva
Presentation: Canva / Google Slides
Image Data: Phone camera / webcam
The workbook’s project specification allows students to choose among Python, no-code ML, Generative AI, Vision, NLP and Sensors, but requires them to justify why they selected that technology and what alternatives they rejected.
Innovator Pitch — Live Demo & Defence
The final presentation is more demanding than in the earlier classes.
Students deliver a five-minute pitch followed by three minutes of questions.
Six Pitch Sections
1. Problem
Introduce the named person who has the problem.
2. Solution
Explain the solution in one sentence.
3. How It Works
Run the working component live.
4. Benefits
Explain how success will be measured.
5. Limits
Present the two failure modes the team discovered.
6. Responsible Use & Next Steps
Explain who should use the system, how AI was used and what would be improved next.
Projects are judged on:
Problem Clarity • AI Usefulness • Technical Understanding • Prototype/Demo • Stated Limits • Creativity • Teamwork • Presentation & Q&A
Skills Students Develop
Through the three-day Class 10 programme, students develop:
AI Systems Understanding • Python Programming • Debugging • Input Validation • Data Cleaning • Data Auditing • Feature Engineering • Data Leakage Awareness • Bias Analysis • Machine Learning Workflow • Train/Validation/Test Evaluation • Classification • Regression • Model Accuracy • Precision & Recall • Overfitting Diagnosis • Generative AI • Advanced Prompt Engineering • AI Verification • Computer Vision • NLP • Chatbot Design • Robotics & IoT • System Specification • Hostile Testing • Failure Analysis • Responsible AI • Prototype Development • Technical Presentation
By the end of the workshop, students are expected to be able to specify a system before building it, test it deliberately, identify where it fails and publicly defend their technical decisions.
Analyse → Evaluate → Engineer → Build → Test → Defend
3 Days. From Understanding AI to Engineering Responsible AI Systems.
The programme concludes with the AI Innovation Challenge, live Innovator Pitch, technical Q&A, Project Exhibition, Grand Finale and Certificate of Completion.
The Class 10 certificate recognises learning across Python, data auditing, machine learning, model evaluation, Generative AI, prompt engineering, computer vision, NLP, robotics & IoT and system design, as well as the successful specification, testing, improvement and defence of an original AI system.
SKK Educational Foundation
AI + Python + Machine Learning + Model Evaluation + Generative AI + System Innovation