AI curriculum

Course outline for Applied Generative AI

Use generative AI to summarize, extract, transform, and draft useful content. Develop prompts, provide relevant context, request structured results, and check what the model produces. Try a text-and-image task and observe a complete tool exchange. Compare task results to choose a model or a simpler non-model approach.

Applied Generative AI Course Objectives

  • Revise prompts and context for different generation and extraction tasks.
  • Request structured output and check its format and substantive correctness.
  • Operate a tool-call exchange and inspect actual tool results.
  • Use text and image input and identify unsupported interpretations.
  • Compare models and choose simple routing or review rules from observed results.

Pre-requisites

  • Artificial Intelligence Foundations or equivalent understanding of AI tasks, learning, and model limitations.
  • Use a browser, documents, and simple tables; read basic JSON with an introductory bridge.
  • How Large Language Models Work is recommended, not required. Programming is optional in this course.

Before this course

Complete the required courses or bring equivalent experience. Recommended courses add useful depth; they are not enrolment gates.

Required preparation

Recommended preparation

Readiness checks

  • Artificial Intelligence Foundations or equivalent understanding of AI tasks, learning, and model limitations.
  • Use a browser, documents, and simple tables; read basic JSON with an introductory bridge.
  • How Large Language Models Work is recommended, not required. Programming is optional in this course.

Understands AI tasks, learning, and model limitations; can use browser documents and simple tables and read basic JSON with an introductory bridge. Programming is optional.

Find your place in the learning map

Lab Setup

Delivery requirements: the following environment, materials, access, and exercise timing must be prepared and verified before delivery. This outline does not certify that a lab package is ready.

A browser-based playground or prepared notebook supporting actual model calls, JSON/schema controls, image input, and a harmless tool exchange. At least one verified multimodal model and two comparable configurations are needed. The instructor must arrange access and quotas. Learners need not implement the tool dispatcher or purchase an account; Building Applications with LLMs teaches application implementation.

Detailed Course Outline

Module 1: Prompting for useful results (2 hours)

  • Task and output instructions: Specify task instructions, examples, constraints, and output requirements for useful generation.
  • Source-based generation: Summarization, extraction, rewriting, and drafting with source material.
  • Prompt examples and revision: Zero-shot and few-shot prompts and systematic prompt revision.
  • Context and sensitive data: Relevant context, examples, token limits, and sensitive-data exclusions.
  • Controlled prompt experiments: Prompt instructions versus model settings; changing one factor at a time.

Practical outcome: Improve prompts for a small document task and explain the observed changes.

Module 2: Structured, multimodal, and tool-assisted tasks (2 hours)

  • JSON and semantic checks: JSON output, schemas, allowed values, and semantic checks.
  • Schema-constrained output: Compare prompted JSON with schema-constrained generation where the selected model supports it.
  • Text and image input: Use text-and-image requests and check extracted visual facts against their source.
  • Tool requests: Tool descriptions, argument schemas, and model-selected function calls.
  • Executed tool results: Application execution, tool results, and a follow-up model response.

Practical outcome: Produce checked structured data and complete a harmless read-only tool exchange.

Module 3: Selecting models and handling difficult inputs (2 hours)

  • Task checks and baselines: Task criteria, simple baselines, and separate development and check cases.
  • Model selection criteria: Model quality, latency, input capabilities, context, and usage cost.
  • Classification alternatives: Classification with a generative model versus rules or a classical classifier.
  • Routing and human review: Explicit routing, abstention, and human review for unsupported cases.
  • Image generation and speech orientation: Brief orientation to image generation and speech tasks; modality-specific constraints.

Practical outcome: Compare two available model configurations and justify a task-specific choice.

Practical exercise

  • Use a small event-information pack to summarize an update, extract facts into JSON, and draft an audience-specific message. Check an event-poster image against the extracted facts. Let a model request a harmless schedule lookup through a prepared tool interface, execute the lookup, and return its result to the model. Compare revised prompts and two configurations on a few unseen examples.

What you will produce

Revise prompts, check structured and multimodal results, complete an executed tool exchange, and justify a task-specific model choice.

How the work is checked

  • Prompt changes address observed failures rather than add arbitrary wording.
  • Format validity and source-supported content are checked separately.
  • The learner distinguishes a requested tool call from its observed execution result.
  • A new image or document example is checked for missing or invented facts.
  • The model or routing choice includes evidence and a simpler alternative.

Six guided hours include demonstrations, practice, feedback, and questions. Breaks and pre-course setup are outside the total. Core tasks include text output, image understanding, and one executed tool exchange. Image generation and speech receive orientation only. This is applied interaction, not a full multimodal-production or agent course. Model training, fine-tuning, RAG pipelines, durable app state, and production routing infrastructure are outside scope. Evaluation is a small task check; AI Application Evaluation develops a systematic program. Access, quotas, and the selected multimodal/tool path require verification before delivery.

When another approach is better

  • Use rules or a classical classifier when they meet the task with simpler checks.
  • Use Building Applications with LLMs to implement the tool dispatcher and application state.

How we train

Contact us for full course details, including duration, delivery options and lab requirements.

Ask about exercises, instructor feedback and the prior knowledge you need. Tool-specific courses marked provisional may become modules in a broader course.

Discuss your learning goals