Tools and Emerging Technologies
Domain 9 - Tools and Emerging Technologies. This domain tests whether you know what each tool in the AI stack actually does, and can assemble a sensible toolchain for a stated task.
Core rules and concepts
- Environment: Python is the default language; Jupyter Notebook mixes code, output and explanation; Google Colab provides free GPU access; Git and GitHub handle version control; VS Code suits scripts.
- Libraries by job: NumPy for arrays and maths; pandas for tabular loading, cleaning and grouping; Matplotlib and Seaborn for charts; scikit-learn for classical ML (regression, classification, clustering, preprocessing, metrics); TensorFlow/Keras and PyTorch for deep learning; NLTK and spaCy for NLP; OpenCV for image handling; joblib or pickle for saving models.
- Deployment: serialise the model and serve it behind an API (Flask, FastAPI), or call a cloud AI service (vision, speech, language) when you lack data or expertise; no-code tools such as Teachable Machine suit quick prototypes.
- Emerging technologies: large language models and retrieval-augmented generation, edge AI running on devices, IoT sensor streams, digital twins, federated learning for privacy, and AutoML.
Worked example
Build a sentiment dashboard for school feedback. pandas loads and cleans the CSV; spaCy or NLTK tokenises the text; scikit-learn TfidfVectorizer plus LogisticRegression trains the classifier; matplotlib draws the class distribution and monthly trend; joblib saves the model; Jupyter documents the whole run so it can be reproduced. Each choice is justified by the function it performs in the pipeline, not by popularity.
Traps that cost marks
- Saying Matplotlib trains models, or that pandas performs the classification.
- Using scikit-learn when the task requires a large convolutional network; deep learning frameworks are the right tool there.
- Assuming a cloud API removes responsibility for consent and data protection.
- Confusing a NumPy array with a pandas DataFrame when labelled columns are needed.