Planning and designing an AI solution

Your school wants to reduce the amount of recyclable waste that goes into general bins. You have been asked to build an AI system that looks at an item through a webcam and tells the user whether it goes in recycling, general waste or food waste. Requirements: - the system must classify an item held up to a camera into one of the three classes - it must show its confidence for each answer - it must be tested on items that were not used to train it - you must evaluate how well it would work in the school canteen Suggested tool: Teachable Machine (image project), or a similar beginner image-classification tool.

Year 10 students often do not know how to plan their revision. Build a rule-based expert system that asks a student a few questions and gives personalised revision advice. Requirements: - the system must ask at least four questions, e.g. weeks until the exam, hours available per week, confidence in the subject (1–5), preferred way of learning - it must use at least eight IF–THEN rules - it must give a clear piece of advice and explain why it gave that advice - it must deal sensibly with unexpected answers Suggested tools: Python, Scratch, or a decision tree drawn and traced by hand.

A company is designing a robot floor cleaner. Before building the real robot, you must create an autonomous routine for a virtual robot in a simulated room. The room is a 6 × 6 grid. Some squares contain furniture (obstacles). The robot can sense whether the square ahead is blocked, whether the current square is dirty, and its battery level. Requirements: - the robot must clean every reachable dirty square without human control - it must avoid obstacles - it must return to its charging dock (square 1,1) when the battery is at 20% or below - each move uses 2% battery; each clean uses 1% Suggested tools: Scratch, Python (e.g. a grid printed in the console), or a hand-traced grid on paper.

A local sports club wants to know the day before whether outdoor training is likely to be rained off. Build a machine learning model that predicts Rain tomorrow: Yes/No from today's weather. Starter data (you may collect more):

Day	Humidity (%)	Pressure (hPa)	Cloud cover (0–8)	Wind (km/h)	Rain tomorrow
1	85	1002	7	22	Yes
2	60	1021	2	10	No
3	90	998	8	30	Yes
4	55	1025	1	8	No
5	78	1008	6	18	Yes
6	65	1018	3	12	No
7		1004	7	25	Yes
8	50	1030	0	5	No
9	88	1000	8	28	Yes
10	70	1015	4	14	No
11	82	1006	6	20	Yes
12	58	1023	2	9	No
13	145	1012	5	16	No
14	75	1010	5	17	Yes
15	62	1020	3	11	No
16	60	1021	2	10	No

Requirements: - prepare the data, then train a classification model (e.g. a decision tree) - test it on data it was not trained on - explain how the club should use the prediction Suggested tools: Python (e.g. scikit-learn), a spreadsheet, or a decision tree built and traced by hand.

The school canteen wants to understand its customers so it can plan menus and offers. Use clustering to find groups of students with similar buying habits. Starter data for one month (anonymised):

Student	Visits per week	Average spend (£)	% of purchases that are hot meals
S1	5	2.10	10
S2	5	4.80	90
S3	1	1.50	0
S4	4	4.50	85
S5	2	1.20	5
S6	5	2.40	15
S7	1	1.80	10
S8	5	5.10	95
S9	4	2.00	20
S10	2	1.60	0
S11	4	4.90	80
S12	5	2.20	10

Requirements: - find a sensible number of clusters and describe each group - explain how the canteen could use each group - consider privacy, because this is data about real students Suggested tools: Python (e.g. k-means), a spreadsheet scatter graph, or grouping by hand from a plotted graph.

A small online shop receives hundreds of customer reviews and wants to spot unhappy customers quickly. Build an AI system that classifies a written review as positive or negative. Starter reviews:

Review	Label
"Arrived quickly and works perfectly"	positive
"Broke after two days, really disappointed"	negative
"Great value, would buy again"	positive
"Terrible customer service, never again"	negative
"Love it! Exactly as described"	positive
"The size was wrong and the refund took weeks"	negative
"Not bad at all, actually really good"	positive
"Oh great, another charger that doesn't charge"	negative

Requirements: - collect at least 40 labelled reviews in total (you may write realistic ones) - train and test a classifier, or build a keyword-scoring system and compare it with a trained model - flag negative reviews for a human to reply to Suggested tools: Teachable Machine is not suitable for text – use Python, a beginner text-classification tool, or a keyword-scoring spreadsheet.