AI Designer
Design clear, inclusive, and trustworthy AI experiences
A design path for AI experiences: research a service task, establish bilingual RTL and accessibility requirements, design uncertainty and user control, prototype conversational or multimodal flows, evaluate them with users, and prepare an implementation handover.
Prerequisites
- Basic UX/UI knowledge, including task flows, wireframes, interaction states, and usability principles
- Working proficiency with a prototyping tool and interactive components, or completion of the mandatory prototyping prework
- Access to Arabic and English content review and accessibility review support; course-provided reviewers may fulfill this requirement
- Willingness to test designs with diverse users and revise assumptions from evidence
Designing for AI Foundations
Understand how AI behavior changes user experience, trust, control, and the designer’s responsibilities.
- Map an AI experience from user intent through data, model or rules, interface, human decision, feedback, fallback, and operating owner
- Write experience principles that turn uncertainty, variability, evidence, disclosure, control, accessibility, privacy, and failure into observable design commitments
- Define a boundary brief with supported tasks, prohibited decisions, user expectations, human authority, safe alternatives, data limits, and design-owner responsibilities
AI UX Principles in Practice
Apply clear expectations, meaningful evidence, correction, fallback, and user control to an AI experience.
AI Design Foundations
Distinguish deterministic interface behavior from probabilistic AI behavior and design each honestly.
Set AI Experience Boundaries
Define what the experience helps with, what it cannot do, when it asks for input, and when it hands off.
Lab assessment
Choose one bounded service experience and define its design contract before creating a conversational flow or polished prototype. Compare the AI concept with a non-AI alternative.
Submit the experience-system map, AI-versus-non-AI comparison, principle set with examples, boundary brief, failure scenarios, responsibility map, and bilingual summary.
The map correctly connects user intent, data, probabilistic or deterministic behavior, interface states, human decisions, operations, feedback, and fallback with named owners.
Principles are specific, non-conflicting, testable through interface behavior, and address utility, evidence, uncertainty, control, accessibility, privacy, and failure.
Supported and prohibited tasks, data use, user expectations, human authority, fallback, escalation, and design and operational ownership are explicit.
The brief traces principles to user or service evidence, compares a credible non-AI path, exposes assumptions, and identifies conditions requiring redesign or no AI.
Arabic and English principles and boundaries preserve meaning, avoid misleading anthropomorphism, and use consistent terms for uncertainty, control, and responsibility.
User Research and Task Design
Research real service tasks and translate evidence into flows that respect different users and contexts.
- Collect authorized evidence from representative Arabic- and English-speaking users, including diverse abilities, contexts, workarounds, exceptions, and service channels
- Synthesize tasks, goals, mental models, language needs, pain points, unequal impacts, evidence strength, contradictions, and open questions
- Produce an evidence-linked task flow and service blueprint showing user actions, visible and backstage steps, AI boundaries, human handoffs, failures, and recovery
Research AI User Needs
Plan research that uncovers user goals, context, trust expectations, correction needs, and harm concerns.
Task Flows and Service Blueprints
Map the user journey, backstage work, data, decisions, handoffs, and failure paths before designing an AI interface.
Design for Diverse Users
Account for language, ability, device, confidence, urgency, and service context rather than designing for one ideal user.
Lab assessment
Use approved course research or authorized participants with appropriate safeguards. Include people who differ in language, ability, confidence, channel, and service context.
Submit the research plan and safeguards, participant matrix, de-identified notes, evidence repository, synthesis, priority task statement, annotated task flow, service blueprint, and research limitations.
The sample and method fit the question, sources are traceable, observations and quotations are separated from interpretation, and contradictory evidence is retained.
Findings cover goals, behavior, language, ability, channel, context, workarounds, exceptions, unequal effects, evidence strength, and gaps without inventing a universal user.
The flow follows research evidence and includes triggers, user decisions, inputs, outputs, uncertainty, errors, corrections, handoffs, fallback, completion, and abandonment.
Visible and backstage actions, people, systems, data, policies, evidence, AI boundaries, dependencies, failure recovery, and ownership align across the service.
Consent, privacy, de-identification, access, retention, translation review, and accessibility are documented, and Arabic evidence is not distorted by English synthesis.
Bilingual RTL Accessibility
Create Arabic and English AI experiences that are readable, culturally appropriate, and accessible from the first prototype.
- Specify Arabic RTL and English LTR content, layout, navigation, mirroring, mixed-direction text, numbers, dates, icons, and language-switch behavior for the researched task
- Define testable accessibility requirements for semantics, keyboard, focus, reading order, labels, contrast, zoom, motion, errors, status messages, and assistive technology
- Complete bilingual content and accessibility reviews, resolve critical findings, and record accepted design debt with owners and verification criteria
Design RTL and Bilingual Content
Design language switching, direction, labels, terminology, dates, numbers, and mixed-language identifiers intentionally.
Design Accessible AI Interfaces
Ensure people can perceive, understand, navigate, and correct AI interactions across different devices and abilities.
Localize Cultural and Service Context
Validate examples, tone, expectations, explanations, and escalation guidance with the people and service context they serve.
Lab assessment
Complete this specification before trust patterns, interaction flows, or prototype production. Apply it to representative task screens and have both language and accessibility reviewers inspect the result.
Submit the bilingual content model, RTL and LTR layout rules, component and state requirements, accessibility acceptance checklist, representative annotated screens, reviewer findings, resolution log, and design-debt register.
Content preserves task meaning, information hierarchy, uncertainty, warnings, evidence, control, error recovery, and tone across languages without literal mistranslation.
Layout, reading and focus order, navigation, mirroring exceptions, alignment, mixed scripts, numbers, dates, punctuation, and icons behave correctly in RTL and LTR.
Requirements are observable and cover structure, names and roles, keyboard, focus, contrast, zoom, reflow, motion, errors, status, timing, and assistive technology.
Named language and accessibility reviewers test representative states, findings include severity and evidence, critical issues are resolved, and fixes are rechecked.
Unsupported contexts, reviewer limits, unresolved issues, impact, workaround, owner, due date, and acceptance authority are explicit.
Trust, Uncertainty, and Control
Design interactions that show what AI knows, does not know, and how a user can remain in control.
- Design disclosure and evidence patterns that identify AI involvement, source or rationale, freshness, confidence limits, and the responsible human or service
- Prototype calibrated patterns for uncertain, conflicting, incomplete, unsupported, stale, unsafe, unavailable, and corrected outcomes without false confidence
- Specify meaningful user controls to inspect, edit, retry, refuse, undo, switch to a manual path, escalate, complain, appeal, and stop
Design for Trust, Not Blind Confidence
Use predictable behavior, source visibility, clear scope, and honest language to earn appropriate trust.
Communicate Uncertainty Clearly
Show confidence, missing evidence, ambiguity, and limitations in ways that help users choose a safe next action.
Give Users Control and Correction
Design edit, retry, reject, report, undo, handoff, and confirmation patterns around AI output.
Lab assessment
Create reusable patterns for the researched task using the approved bilingual and accessibility specification. Test the patterns as complete decision moments, not isolated visual components.
Submit the pattern inventory, annotated Arabic and English states, trigger and content rules, interaction prototypes, evidence map, scenario test results, and usage and anti-pattern guidance.
Patterns make AI involvement, purpose, evidence, freshness, limits, data use, and accountable service visible at the point they affect user judgment.
Uncertain, conflicting, incomplete, unsupported, stale, unsafe, unavailable, and corrected states use distinct truthful content and lead to safe next actions.
Controls are available before consequence, preserve entered work, provide correction and undo where feasible, and offer human, manual, complaint, appeal, and stop routes.
Representative scenario tests show comprehension and safe action, expose over-trust and under-trust risks, and document unsupported contexts and residual design debt.
Arabic RTL and English LTR patterns preserve meaning, prominence, order, warning strength, keyboard access, focus behavior, and assistive labels.
Conversational and Multimodal Flows
Design conversational, visual, voice, and tool-assisted experiences that keep tasks and boundaries clear.
- Model the complete conversation with entry, intent clarification, context collection, confirmation, evidence, uncertainty, correction, completion, interruption, return, and exit states
- Coordinate text, visual, voice, upload, and tool-assisted steps with clear modality choice, equivalent alternatives, handoffs, permissions, and state continuity
- Prototype tool and agent boundaries that preview arguments and consequences, request informed approval, expose status, handle partial failure, and support cancellation and recovery
Design a Conversational Flow
Structure a conversation with purpose, clarification, confirmation, evidence, repair, and a clear end state.
Design Multimodal Interaction
Choose text, voice, image, document, and form inputs according to user needs, accessibility, and data sensitivity.
Design Tool and Agent Experiences
Make tool actions, progress, permission, confirmation, errors, and human oversight visible when AI can take a step.
Lab assessment
Apply the bilingual, accessibility, trust, and control specifications to one complete task. Use the simplest modalities needed and keep all irreversible actions outside the prototype.
Submit the state and conversation model, Arabic and English scripts, multimodal storyboard, tool or agent approval flow, interactive prototype, scenario results, and edge-case and limitation log.
The model covers entry through exit, preserves task context, prevents loops and dead ends, and handles clarification, correction, interruption, return, failure, and abandonment.
Each modality has a justified role, information and state remain synchronized, users can switch or use an equivalent alternative, and permissions and errors are clear.
Purpose, arguments, target, consequence, uncertainty, status, approval, cancellation, partial failure, recovery, and audit are visible before or during tool use.
Tests cover normal, ambiguous, conflicting, unsupported, inaccessible, permission-denied, interrupted, failed-tool, and recovery paths and document residual risks.
Arabic and English flows preserve intent, turn order, controls, evidence, warnings, fallback, voice alternatives, captions, and accessible labels.
Generative AI Design Workflow
Use generative tools to explore design options while preserving research evidence, originality, safety, and review.
- Prepare bounded design prompts from approved research, principles, bilingual and accessibility specifications, trust patterns, and task constraints without including prohibited data
- Generate and curate genuinely different design options while recording tool, model, inputs, outputs, edits, provenance, rights, and rejected directions
- Select, revise, or reject options through human review against user evidence, task success, originality, accessibility, bilingual quality, safety, feasibility, and known limits
Use a Generative Design Workflow
Generate options from a clear brief, compare them against user evidence, and preserve the rationale for chosen directions.
Prompt for Design Exploration
Give generative tools a safe brief with user task, constraints, accessibility, language, and criteria rather than vague style requests.
Review Generated Design Output
Inspect generated visuals, copy, flows, accessibility, cultural fit, licensing, and unintended assumptions before reuse.
Lab assessment
Use only approved generative tools and non-sensitive project context. Treat outputs as untrusted proposals, preserve human authorship and decision responsibility, and retain the complete review trail.
Submit the exploration brief, prompt and constraint set, unedited output sample, curated options, provenance and rights record, completed review matrix, revised selected concept, rejection reasons, and decision log.
Prompts trace to approved research and specifications, define task and exclusions, avoid prohibited data, and do not ask the tool to invent user evidence.
Options differ in meaningful interaction choices, and the log records tool, model, input, output, edits, source material, rights, and rejected branches.
Reviewers verify task fit, evidence, hierarchy, states, interaction, feasibility, originality, and implementation constraints rather than selecting visual polish alone.
The record addresses privacy, confidential context, bias, stereotypes, deceptive patterns, intellectual-property uncertainty, tool terms, unsupported claims, and residual risk.
Arabic RTL and English LTR concepts are reviewed for content parity, typography, mixed direction, reading and focus order, contrast, labels, errors, and assistive use.
Prototype Evaluation and Iteration
Test AI experience prototypes with users, measure task success and trust, and turn evidence into focused improvements.
- Run moderated task evaluations with representative Arabic- and English-speaking users across normal, uncertain, error, fallback, correction, and accessibility scenarios
- Measure task completion, error recovery, time or effort, comprehension, evidence use, calibrated reliance, control use, accessibility, and qualitative findings
- Prioritize an iteration backlog by user and safety impact, evidence strength, affected groups, effort, dependency, owner, verification criterion, and accepted design debt
Test AI Prototypes with Users
Plan task-based sessions that reveal understanding, trust, correction, error recovery, and real user behavior.
Evaluate Trust and Task Success
Measure whether users complete the task, know the system’s limits, notice evidence, and choose safe next actions.
Iterate and Manage Design Debt
Prioritize fixes that reduce harm or confusion, document known design debt, and avoid polishing around a broken task flow.
Lab assessment
Evaluate a testable prototype with approved participants or course sessions. Separate observed behavior, participant statement, facilitator interpretation, metric, and design decision.
Submit the test plan and safeguards, participant matrix, scenarios and success criteria, de-identified notes and recordings where approved, metric results, finding-to-evidence table, evaluation report, prioritized backlog, and retest plan.
Participants, tasks, languages, abilities, contexts, scenarios, success criteria, facilitation, and safeguards match the research question and known risks.
Results measure completion, errors, recovery, effort, comprehension, evidence, reliance, control, accessibility, and confidence without treating stated preference as behavior.
Each finding traces to observations and measures, reports frequency and severity where appropriate, distinguishes patterns from anecdotes, and retains contradictory evidence.
Backlog priority follows user and safety impact plus evidence, fixes have owners and verification, and unresolved limitations and design debt are explicit.
The report compares Arabic RTL and English LTR task performance and accessibility without masking language-specific or assistive-technology failures in aggregate results.
Responsible AI Design Capstone
Deliver a bilingual, accessible, trustworthy AI experience with evidence, limits, and a clear handover.
- Deliver a testable Arabic RTL and English LTR prototype that implements the researched task, accessibility specification, uncertainty patterns, user controls, multimodal boundaries, and safe fallback
- Trace each major design decision to user evidence, service constraints, evaluation findings, responsible-AI controls, considered alternatives, and accepted limitations
- Hand over flows, content, components, states, design tokens, interaction rules, accessibility behavior, assets, test evidence, open debt, owners, and implementation acceptance criteria
Write the Responsible AI Design Brief
Combine research, task, boundaries, accessibility, trust patterns, evaluation, and handoff requirements in one brief.
Build the Capstone Prototype and Handover
Create a testable prototype with documented language, accessibility, error, uncertainty, and escalation behavior.
Review the Design Capstone
Review the experience with users and stakeholders against task success, safety, inclusion, and implementation readiness.
Lab assessment
Integrate and resolve the preceding artifacts into one implementation-ready handover. Do not conceal failed evaluation findings with visual polish or claim accessibility, trust, or readiness beyond recorded evidence.
Submit the responsible design brief, interactive bilingual prototype, indexed specifications and assets, content inventory, design system, annotated flows and states, evaluation and review evidence, decision log, limitation and debt register, and signed implementation handover.
The prototype supports the researched task through entry, evidence, uncertainty, control, completion, error, correction, fallback, escalation, and exit without critical dead ends.
Major patterns, content, flows, controls, and exclusions trace to research, constraints, principles, evaluation findings, alternatives, and named decision owners.
AI involvement, evidence, uncertainty, data use, human authority, user correction, approval, complaint, appeal, fallback, prohibited use, and residual risk are designed explicitly.
Arabic RTL and English LTR experiences have reviewed content parity and pass documented keyboard, focus, semantics, contrast, zoom, reflow, error, status, and assistive-technology checks.
A receiving team can locate and implement versioned flows, content, components, states, assets, rules, breakpoints, acceptance criteria, open debt, and owners without hidden verbal context.