IBM Artificial Intelligence

Machine learning finds patterns in examples instead of following written rules. You learn features and labels, the three main learning styles, and the difference between training a model and using it.

• Learning from examples instead of rules
• Features, labels and datasets
• Supervised, unsupervised and reinforcement learning
• Classification and regression
• Training against inference
• Why data quality sets the ceiling

You build A small classifier trained on a public dataset, with each step explained.

Deep learning drives most modern AI. You see how layers of weighted connections learn, why depth and data matter, what the common network families do, and why GPUs became essential.

• Neurons, weights and layers
• How a network learns from its errors
• Why depth helps and what it costs
• Convolutional and recurrent networks in outline
• Hardware: why GPUs
• Machine learning compared with deep learning

You build A simple neural network trained on a toy problem, with its learning curve explained.

A model that looks good in training may fail in use. You learn how data is split, how overfitting happens, which metrics suit which problem, and why bias in data becomes bias in a model.

• Training, validation and test data
• Overfitting and underfitting
• Accuracy, precision and recall
• Reading a confusion matrix
• Data drift after deployment
• How bias enters through data

You build An evaluation of two models that explains which one to trust and why.

Transformers changed what models can do with language. You learn how attention lets a model weigh context, how large models are pretrained, and why they can be adapted to new tasks by prompting or fine-tuning.

• From sequence models to transformers
• Attention in plain language
• Pretraining on large text collections
• Foundation models and their adaptation
• Prompting against fine-tuning
• Context windows and their limits

You build A short explainer of how a large language model produces one sentence.

Language models work on numbers, not words. You see how text becomes tokens, how embeddings place meaning in a vector space, and how vector databases find similar items quickly.

• Tokenization and why token counts matter
• Embeddings as coordinates of meaning
• Measuring similarity between vectors
• What a vector database stores
• Semantic search compared with keyword search
• Choosing what to embed

You build A semantic search over a small document set using embeddings.

Models invent answers when they lack facts. You learn how retrieval grounds a model in your own documents, and how agents combine a model with tools and a loop so that it can act, along with where both still fail.

• Hallucination and why it happens
• The retrieval-augmented generation flow
• Grounding answers in cited sources
• What an AI agent is
• Tools, memory and the agent loop
• Where agents and retrieval still go wrong

You build A question-answering sketch grounded in a short set of documents.

Not every problem needs a generative model. You survey what text, image, code and speech models do well, judge fit, cost and risk for a proposed use case, and plan how to know whether it works.

• Text, image, code and speech generation
• Summarization, drafting and extraction
• Matching a model to a task
• Cost, latency and quality trade-offs
• Evaluating a generative use case
• When not to use generative AI

You build A fit assessment for three proposed generative AI use cases.

AI systems decide things about people. You learn where unfairness comes from, how transparency and explainability differ, what accountability looks like inside an organization, and how privacy limits what data may be used.

• Sources of bias in data and design
• Fairness and how it can be measured
• Transparency against explainability
• Privacy and consent
• Accountability for automated decisions
• Questions to ask before deploying a system

You build An ethics review of one AI use case, with risks and mitigations written out.

AI adds new ways to be attacked and new rules to follow. You study prompt injection, data leakage and poisoning, then see how risk-based regulation and governance frameworks organize the response.

• Prompt injection and data leakage
• Training data poisoning and model theft
• Misuse: deepfakes and automated fraud
• Risk-based regulation such as the EU AI Act
• Governance frameworks and management standards
• Policies, documentation and human oversight

You build A threat and control table for an AI assistant that handles customer data.

Take one proposed AI project and assess it across technology, data, ethics, security and regulation. You write the briefing a decision-maker needs, in plain language that holds up to technical questions.

• Choosing and describing the project
• Assessing the technology and data
• Assessing ethical and security risk
• Mapping regulatory duties
• Writing a recommendation

You build A decision briefing on one AI project covering feasibility, risk and governance.