B8 · Computational Thinking and the AI Project Cycle

Overview of the AI Project Cycle

The AI project cycle is a step-by-step framework for building an AI solution. Its stages are problem scoping, data acquisition, data exploration, modelling, evaluation and deployment. It is called a cycle because lessons from later stages often send the team back to improve earlier ones.

8 min Beginner Jump to the activity

What you need: Nothing. This page starts from the beginning, with no programming and no formulas.

Definition

A team wants to build an AI that warns farmers about crop disease. Should they start by collecting thousands of leaf photos? Not yet. First they need to be sure what problem they are solving, and for whom.

The AI project cycle is a step-by-step framework for building an AI solution. Its stages are problem scoping, data acquisition, data exploration, modelling, evaluation and deployment. It is called a cycle because lessons from later stages often send the team back to improve earlier ones.

In simple words

Building an AI is like cooking a new dish for guests. First decide what they would like (scoping), buy the ingredients (data), check they are fresh (exploration), cook (modelling), taste it (evaluation), serve it (deployment), and then listen to the guests and improve the recipe.

Why it matters

Most AI projects that fail do so not because of the technology but because the problem was unclear, the data was poor or the people affected were never asked. Following the cycle helps teams avoid these traps and build something genuinely useful.

Key terms

AI project cycle
The stages of building an AI solution, from defining the problem to using it in the real world.
Problem scoping
Understanding and clearly defining the problem, who has it and why it matters.
Data acquisition
Collecting the data needed to solve the problem.
Data exploration
Studying and cleaning the data to understand it and find problems.
Modelling
Choosing and building the method, rule-based or learning-based, that will make decisions.
Evaluation and deployment
Testing how well the model works, then putting it into real use and monitoring it.

Explained in detail

The six stages at a glance

StageKey questionCrop disease example
Problem scopingWhat is the problem, for whom, and why?Small farmers lose crops because disease is spotted too late
Data acquisitionWhat data do we need, and where can we get it fairly?Photos of healthy and diseased leaves, labelled by experts, with farmers' consent
Data explorationIs the data complete, balanced and correct?Too few photos of one disease; some blurry photos removed
ModellingRules or learning? Which method?A learning model trained on the labelled photos
EvaluationHow well does it work, and for whom?90 per cent correct overall, but weak on photos taken at dusk
DeploymentHow will people use it safely?A phone app in local languages, with advice to confirm with an expert

Why it is a cycle

In practice, teams rarely go straight through. Evaluation might show the model is poor on photos taken in evening light, which sends the team back to collect more such photos. Deployment might reveal that farmers cannot read long messages, which sends the team back to problem scoping to rethink the design. Each loop around the cycle improves the solution. Good teams expect this and plan time for it.

Ethics runs through every stage

  • Scoping: does this problem need AI at all? Who might be harmed?
  • Data: was it collected with consent? Does it represent everyone affected?
  • Modelling and evaluation: is it fair across groups? Can its decisions be explained?
  • Deployment: what happens when it is wrong? Who is responsible? How can people complain?

You do not always need AI

One of the most valuable outcomes of problem scoping is discovering that AI is not the right tool. If the crop problem were really that farmers had no access to affordable pesticides, an app would not help. If a simple checklist or a phone helpline would solve the problem, it may be cheaper, fairer and easier to trust than an AI. The cycle starts with the problem, not the technology, so that the solution fits the people it is meant for. The next pages explore each stage in detail.

How it works

  1. Problem scoping: define the problem with the people who have it.
  2. Data acquisition: collect the right data fairly and with consent.
  3. Data exploration: clean and study the data.
  4. Modelling: choose rules or learning and build the model.
  5. Evaluation: test on new data and for different groups.
  6. Deployment: release carefully, monitor and improve.

Try it yourself

Around the cycle with a crop app

A student team builds an AI to spot tomato leaf disease.

Order the AI project cycle

Put the six stages in order.

  1. 1Deployment
  2. 2Modelling
  3. 3Evaluation
  4. 4Problem scoping
  5. 5Data acquisition
  6. 6Data exploration

Common mistakes

  • Jumping straight to building a model before understanding the problem.
  • Treating the cycle as a straight line and never going back.
  • Leaving ethics until the end instead of thinking about it at every stage.

Check yourself

Question 1

1. Which stage comes first in the AI project cycle?

Question 2

2. Testing shows the model fails on evening photos. What should the team do?

Question 3

3. What might problem scoping reveal?

Summary

  • The AI project cycle: scoping, data acquisition, exploration, modelling, evaluation, deployment.
  • It loops back when later stages reveal problems.
  • Ethics belongs at every stage, and sometimes AI is not the answer.

Finished reading?

Sources: CBSE Artificial Intelligence curriculum resources · IndiaAI (Government of India)