The AI Project Cycle

Hello students, welcome to Kwickprep. You would not build a house without a plan. Building an AI system also needs a plan, and that plan is called the AI project cycle. Today we learn its five stages, walk one real school problem through every stage, and see where deployment fits.

First, what is a project cycle? It is a fixed order of steps we follow to build an AI solution. The output of each stage becomes the input of the next stage. It is called a cycle because, after checking the result, we can go back to the start and improve.

Here are the five stages drawn in order. Stage one is problem scoping, where we understand the problem. Stage two is data acquisition, where we collect data. Stage three is data exploration, where we study the data. Stage four is modelling, where we build the AI model. Stage five is evaluation, where we test the model. If the result is not good enough, the arrow loops back, and we go round the cycle again.

Each stage answers one key question. Problem scoping asks, what exactly are we solving, and for whom? Data acquisition asks, what data do we need, and where will it come from? Data exploration asks, what patterns does the data show? Modelling asks, which model or method should we use? Evaluation asks, is the model accurate and good enough to use?

Let us look closer at each stage. Problem scoping means understanding the problem before touching any data. We ask who has the problem, what it is, where it happens, and why solving it matters. At the end, we write one clear problem statement.

Stage two is data acquisition, which simply means collecting data. The data must be correct and must relate to the problem. It can come from surveys, sensors, cameras, old records or trusted websites. We keep part of it as training data to teach the model, and part as testing data to check it later.

Stage three is data exploration. First, we clean the data by fixing errors and missing values. Then we draw charts and graphs, because a picture shows patterns faster than a table. From these charts, we spot patterns and trends that guide the model.

Stage four is modelling, where we build the AI model. In a rule based approach, humans write the rules, like if the day is Friday then cook less. In a learning based approach, the machine learns the rules from the data itself. We then train the chosen model on the training data.

Stage five is evaluation. We test the model on testing data, which it never saw during training. We compare its predictions with the real answers to see how often it is right. If it is good enough, we use it, otherwise we loop back and improve.

Now let us walk one real problem through every stage, food waste in a school canteen. In problem scoping, we find that the canteen throws away too much food, which wastes money. In data acquisition, we collect six months of records on meals cooked, students present, weather and the day. In data exploration, a chart shows that Fridays and rainy days waste the most food. In modelling, we train a model to predict how many plates will be eaten tomorrow. In evaluation, we test it on one month of data it never saw.

Now pause and predict. On a test Friday, the model predicts four hundred plates will be eaten. But only three hundred plates are actually eaten. Which stage do we return to? Evaluation shows the error, so we loop back, perhaps to add exam day data in data acquisition, and train again.

Finally, where does deployment fit? Deployment means putting the tested model into real use, like a simple app the canteen manager checks every morning. It comes only after evaluation shows the model is good enough, and some newer textbooks show deployment as a sixth stage. After deployment, we keep monitoring the model, and when results slip, the cycle starts again.

Let us recap. The AI project cycle is a fixed order of stages that loops back to improve. The first three stages are problem scoping, data acquisition and data exploration. Then come modelling and evaluation. And deployment puts the model into real use, only after it passes evaluation.