Advanced & Multimodal Prompting

Course at a glance

Item	Detail
Title	Advanced & Multimodal Prompting
Level	Intermediate to advanced
Format	Online, self-paced; short lessons, worked examples, hands-on labs
Duration	~10 hours (about 75–90 minutes per unit)
Audience	Professionals who already use AI assistants and want consistent, production-quality results: analysts, engineers, coordinators, trainers, managers, content and HR teams
Prerequisites	Regular use of an AI assistant (e.g. Claude, ChatGPT, Gemini); comfort with basic prompting (clear instructions, role, examples). No coding required; optional extensions for technical learners
Tools	Any current multimodal AI assistant that accepts files, images and audio/video; a spreadsheet or text editor
Assessment	Self-assessed: knowledge checks after each unit, a rubric-scored practical lab per unit, and a final self-assessment
Completion	Pass all unit knowledge checks (80%+) and the final self-assessment (80%+)

 

Course description

Most people get decent answers from AI by typing a quick request. Getting reliable, repeatable, high-quality results is a different skill. This course teaches it.

Advanced & Multimodal Prompting goes well beyond the basics. You will learn to design the context a model works from, so it understands your goal, audience, constraints and source material. You will learn to request structured outputs, such as tables, JSON and fixed templates, that drop straight into reports, spreadsheets and systems. You will break complex jobs into prompt chains, where each step feeds the next and can be checked along the way.

The course then moves past text. You will work with images, scanned and digital documents, charts, audio recordings and video as inputs. You will extract data from a photographed form, summarise a long PDF with citations, turn a meeting recording into actions and analyse a site walkthrough video. Throughout, you will learn where these models go wrong and how to verify, test and improve your prompts before you rely on them.

Each unit combines short lessons with worked examples and a hands-on lab you score against a clear rubric. You finish with a personal prompt library and a tested, multi-step multimodal workflow you can use at work.

Short catalog blurb (≤ 50 words): Go beyond basic prompting. Learn to design context, get structured outputs, chain prompts into reliable workflows, and use images, documents, audio and video as inputs. Self-paced with hands-on labs, rubrics and a final self-assessment, so you leave with tested prompts and workflows you can use immediately.

Learning outcomes

By the end of this course, learners will be able to:

1. Diagnose why a prompt produces weak, inconsistent or incorrect output, and apply targeted fixes rather than trial and error.
2. Design layered context for a task: role, goal, audience, source material, constraints, success criteria and examples, organised with clear delimiters such as XML-style tags.
3. Specify structured outputs (tables, JSON, fixed templates, schemas) and validate that responses conform to them.
4. Decompose a complex task into a prompt chain with defined inputs, outputs and checkpoints, including self-review and critique steps.
5. Prompt with images, including photos, screenshots, diagrams, charts and handwriting, to describe, extract, compare and reason about visual content.
6. Work with documents such as long PDFs, scanned files, spreadsheets and multi-file sets: extract data, summarise with citations and answer questions grounded in the source.
7. Use audio and video as inputs to transcribe, summarise, extract actions and analyse content, choosing between native media input and transcript-first approaches.
8. Evaluate and verify AI output by grounding answers in sources, detecting hallucinations, and testing prompts against a small set of test cases.
9. Apply responsible-use practices when sharing files and media with AI: data privacy, confidentiality, consent for recordings, and appropriate human review.
10. Build and document a reusable prompt library and an end-to-end multimodal workflow for a real task in their own role.

Course outline

Seven units build in sequence: text techniques first (Units 1–3), then each input type (Units 4–6), then verification and a capstone workflow (Unit 7). Each unit ends with a lab scored against a rubric and a 10-question knowledge check.

Unit	Title	Outcomes
1	Beyond the Basics: How Models Read Your Prompt	1, 8
2	Context Design	2
3	Structured Outputs and Prompt Chaining	3, 4
4	Prompting with Images	5
5	Working with Documents and Data Files	6
6	Audio and Video as Inputs	7, 9
7	Verification, Testing and Your Multimodal Workflow	8, 9, 10

 

Unit 1 — Beyond the Basics: How Models Read Your Prompt

1. What changes between basic and advanced prompting: from one-off requests to designed, repeatable systems
2. How models process a prompt: context windows, tokens, attention to position, and why order and clarity matter
3. Common failure patterns: vague goals, missing context, conflicting instructions, overloaded single prompts
4. A diagnostic checklist for fixing a weak prompt
5. Reasoning techniques: asking for step-by-step thinking, extended thinking modes, and when they help

Lab: Take three weak prompts from a work scenario, diagnose each with the checklist, rewrite and compare outputs side by side.

Unit 2 — Context Design

1. The anatomy of a well-designed context: role, goal, audience, background, constraints, success criteria
2. Structuring long prompts with headings and XML-style tags to separate instructions from data
3. Few-shot examples: choosing, formatting and varying them; avoiding over-fitting to one example
4. Supplying reference material: placement, labelling, and telling the model what to do with it
5. System prompts, custom instructions and project knowledge for reusable context
6. Managing long conversations: when to restart, summarise or re-anchor

Lab: Build a reusable context template for a recurring task in your role and test it on two different inputs.

Unit 3 — Structured Outputs and Prompt Chaining

1. Why structure matters: outputs that feed spreadsheets, reports, LMS imports and other systems
2. Requesting tables, JSON, CSV and fixed templates; defining fields, types and allowed values
3. Prefilling, schemas and validation; handling missing or uncertain data
4. Prompt chaining: breaking a task into steps with clear inputs and outputs
5. Chain patterns: extract → transform → write; draft → critique → revise; classify → route
6. Building checkpoints and self-review into a chain

Lab: Design a three-step chain that turns raw notes into a structured table and then a formatted summary; validate the table against your schema.

Unit 4 — Prompting with Images

1. What vision models do well and badly: text in images, charts, layouts, counting, spatial detail
2. Image prompting techniques: directing attention, asking for description before interpretation, cropping and resolution
3. Extracting data from photos, screenshots, forms, receipts and handwritten notes
4. Reading charts, diagrams and drawings; comparing multiple images
5. Using images as examples or style references

Lab: Extract fields from a photographed form into JSON, then check accuracy against the original and record the error rate.

Unit 5 — Working with Documents and Data Files

1. How models handle PDFs, Word files, slides and scanned documents; text versus visual reading
2. Long-document strategies: placing documents first, quoting before answering, section-by-section processing
3. Summarising with citations and page references; grounding answers in the source
4. Multi-document tasks: comparing versions, cross-referencing policies, consolidating findings
5. Spreadsheets and CSVs: analysis requests, sense-checking figures, avoiding invented numbers

Lab: Summarise a 20+ page policy or report with page-cited key points, then answer five questions and verify each citation.

Unit 6 — Audio and Video as Inputs

1. Native audio and video input versus transcript-first workflows; choosing the right approach
2. Transcription quality: speakers, accents, jargon, noisy recordings
3. Meeting and interview recordings: summaries, decisions, action items with owners and dates
4. Video analysis: timestamps, describing scenes, extracting steps from a demonstration or walkthrough
5. Privacy, consent and confidentiality when uploading recordings

Lab: Turn a recorded meeting or training video into a timestamped summary and an action table; spot-check five items against the recording.

Unit 7 — Verification, Testing and Your Multimodal Workflow

1. Hallucination risks across text, images, documents and media; how to detect them
2. Verification techniques: source quoting, confidence flags, second-pass checking, human review points
3. Testing prompts: building a small test set, comparing versions, tracking results
4. Building and maintaining a personal prompt library: naming, versioning, documenting
5. Designing an end-to-end workflow that combines context design, structured output, chaining and at least one non-text input

Lab (capstone): Build, test and document a multimodal workflow for a real task in your role, scored against the capstone rubric.