We tested the experience before the intelligence behind it was fully connected.
An AI tool that turns ASU faculty's course materials into quiz and exam questions. I helped make its workflow clearer and more controllable, then built the systems that keep that clarity through implementation.
My role Product Designer · one of three designers · ASU LEI and ASU Enterprise Technology
Faculty needed clearer orientation, visible control, and understandable progress before trusting the workflow. The generation services weren't ready to test yet.
Tested a clickable prototype with 16 faculty, logged a decision for every finding, and designed v2 across 6 flows.
v2 is designed and documented. Some items are in development or saved for later, and each one is labelled.
Research evidence, design decisions, and delivery status are labelled separately. Nothing on this page is labelled shipped.
The generative services that write the questions were still being built. Waiting for them would have meant learning about faculty trust only after the expensive part was finished. So we built a research prototype in 3 days while engineering tested feasibility.
Remote moderated sessions, April 28 to May 6, 2026
ASU faculty from math to the humanities
Evaluated task flows, first visit to export
Post-session questionnaire answers from faculty
The research team ran five sessions with 16 ASU faculty. I helped synthesize the evidence and translated it into interface and system priorities.
Moderation: Shailee Shah and Ishrat Ahmed. Report: Shailee Shah, Ishrat Ahmed, and Scotty Craig. I own the design decisions shown here.
These labels summarize the research. Open the section below to read what faculty experienced.
Faculty weren't sure what the tool was for, what a project or question set meant, or what to do next.
"Knowledge" was easy to miss, and the two file roles did not explain how each source affected the AI.
Labels did not match how faculty plan assessments. "Fast vs Slow" was mistaken for difficulty.
Processing messages moved too quickly to read, and the total question count was easy to miss.
Edit and Regenerate were hard to discover, and faculty could not guide regeneration.
Faculty could not tell the formats apart or which QTI—the file format Canvas imports—would work.
Each priority connects what faculty experienced to the change it produced.
50 findings→6 themes→6 v2 flows
Designed in v2
Designed in v2
Designed in v2
Designed in v2
Designed in v2
Designed in v2
Filter the evidenceChoose a status to update the table. Use “High priority only” to narrow it further.
Showing all 50 findings.
Research findings and team-reported responses. Priority comes from the research report; responses and statuses come from the team's action log, dated September 30, 2026.
| Task | Observed | Priority | Team response | Status |
|---|---|---|---|---|
| T1 | Unsure what to do when the dashboard is empty | High | Empty dashboard that leads to the first question set | Designed in v2 |
| T1 | "Create project" didn't signal question creation | Low | Project onboarding flow, reported as coming soon | Reported in development |
| T1 | No next step after naming a set; "View" felt passive | High | Action label and a next-step cue | Designed in v2 |
| T1 | Page purpose unclear | Low | One-line descriptor through onboarding, reported as coming soon | Reported in development |
| T1 | Wanted a role and course setup step | Low | Personalized first-run setup deferred | Future iteration |
| T2 | Scope of a "question set" unclear | Low | Helper text by the name field | Designed in v2 |
| T2 | Question types not visible on this page | High | Question format shown for each type | Designed in v2 |
| T2 | "Knowledge" tab unnoticed and too generic | High | Renamed Course Material | Designed in v2 |
| T2 | A passive nav item looked clickable | Medium | Clear active and current states | Designed in v2 |
| T2 | Unsure whether actions had a required order | Low | Order to be explained in Question Set Instructions | Needs verification |
| T2 | Collapse icon misread as "add files" | Low | Standard collapse icon | Designed in v2 |
| T3 | "Question creation" vs "Context" files confusing | High | Explicit titles, with examples in a popover | Designed in v2 |
| T3 | Context-file tooltip contradicted expectations | High | Tooltip rewritten as an AI usage notice | Designed in v2 |
| T3 | No PowerPoint upload | High | Later iteration | Future iteration |
| T3 | No Google Drive or link upload | Low | Later iteration | Future iteration |
| T3 | Leaving for the Knowledge tab felt like losing work | High | Leave-confirmation check (not auto-save), on the development side | Reported in development |
| T3 | Existing project files mixed with new uploads | Medium | Tooltips plus remove and delete; clear grouping not yet verified | Needs verification |
| T3 | Same helper text repeated three times | Low | Upload area restructured | Designed in v2 |
| T3 | Expected to type instructions, not just upload | High | Prompt input added | Designed in v2 |
| T3 | Copyright concerns not addressed | Medium | Copyright guidance in the upload area | Designed in v2 |
| T4 | Unclear what the configure screen produces | Medium | Copy updated | Designed in v2 |
| T4 | "Fast vs Slow" read as difficulty or speed | Medium | Renamed Generation Quality, with a tooltip | Designed in v2 |
| T4 | Category labels didn't match faculty vocabulary | High | Bloom's removed; question types explained in helper text | Designed in v2 |
| T4 | No difficulty control | High | Not in the current project specs | Needs verification |
| T4 | Question type not selectable before generating | High | Type chosen in configuration and during regeneration | Designed in v2 |
| T4 | Bloom's mentioned with no matching control | Medium | Bloom's removed | Designed in v2 |
| T4 | STEM faculty felt excluded | Low | Later iteration | Future iteration |
| T4 | Leading zero in count fields confused people | Low | Standard number input | Designed in v2 |
| T4 | Framing felt mechanical, not pedagogical | Medium | Goals go in the prompt; type and format stay required | Designed in v2 |
| T5 | Processing screen too fast to read | High | Workflow Details with a completion summary | Designed in v2 |
| T5 | Bloom's label implied an input that didn't exist | Medium | Bloom's removed | Designed in v2 |
| T5 | Expected Bloom's tags on each question | Medium | Resolved by removing Bloom's | Designed in v2 |
| T5 | Total question count missed | Medium | Count may be added to the summary | Future iteration |
| T6 | Edit control not visible | High | Menu removed; controls on each question card | Designed in v2 |
| T6 | Regenerate not a first instinct | High | Menu removed; controls on each question card | Designed in v2 |
| T6 | "Regenerate as [type]" unclear | High | New regeneration flow | Designed in v2 |
| T6 | No way to guide regeneration | High | Regeneration modal: target format plus an optional prompt | Designed in v2 |
| T6 | Correct answer was always the longest option | High | Came from prototype placeholder text; production answer balance not yet verified | Needs verification |
| T6 | Question metadata not visible | High | Question type and format tags; other metadata not verified | Designed in v2 |
| T6 | QTI version unclear for Canvas | Medium | Canvas-compatible versions built by developers | Reported in development |
| T6 | Export options confusing | High | Export copy redesigned | Designed in v2 |
| T6 | No "generate more questions" | High | Later iteration | Future iteration |
| T6 | Way back to configuration not obvious | Low | Question set name in the header | Designed in v2 |
| T6 | Wanted more than four answer options | Medium | Editable options, 2 to 6 | Reported in development |
| T6 | Wanted reordering and answer shuffling | Medium | Shuffling at question and option level | Designed in v2 |
| Gen | Faculty who don't use quizzes saw little value | High | Scenario and visual modes, work in progress | Reported in development |
| Gen | Ethics of faculty using AI for assessment | High | Escalated to project leadership | Leadership decision |
| Gen | Must beat the ChatGPT + QTI workflow | High | Escalated to project leadership | Leadership decision |
| Gen | Wanted memory across past projects | Low | Being added to onboarding | Reported in development |
| Gen | Copy assumed research-faculty materials | Low | Microcopy audited for inclusive examples | Designed in v2 |
Two screens carry the main story: creating a question set and reviewing it.
Workflow DetailsQueued, working, checked, summary ready, on the real frames.Explore Workflow Details →
ExportOne modal, one structured source, five formats.Explore the export system →
Faculty didn't know where to start. The empty state now names the next step.
Designed in v2
Renamed from Knowledge, with two named file roles and an AI usage notice.
Designed in v2
History becomes a tab; the same controls stay visible.
Designed in v2
I owned the tablet and phone layouts. Steps and actions never change order.
Designed in v2
Faculty needed readable progress—not a mysterious spinner.
Try the workflowChoose a state to update the highlighted screen area, or use Play to watch all four.
Summary ready. Every step is checked, a summary says what was made, and View Question set leads on.
Designed in v2 Two real frames cover all four states. No spinner pretending to think very hard.
I structured the copy, components, and history so Figma and code could share one current answer.
Once code could change Figma too, loose text could drift in both directions. Named copy makes mismatches easier to spot.
Everyone can refer to the same interface text.
ASU parts underneath; ReQuesta combinations on top.
One current file and a dated history.
Old and new screens sat together, so developers sometimes built the wrong one. We compared three solutions.
Retired copy stays searchable but separate, so “which one is current?” has one answer.
Designed in v2 Three releases. Figma clips the oldest title: “ReQuesta AI Prototype (Foundational UDS Integration).” No more “final-final-2.”
I tested three directions with a developer and used the structure that fit their workflow.
Desktop and tablet/phone scales sit beside the copy system.
Old copy leaves the active set without breaking files that still use it.
Variable naming grammar: type, then property, then descriptor, joined by underscores.
Used by developers Developers use the named copy in the build; shared names make mismatches easier to catch.
I gave the agent the system rules. It helped produce the work; I reviewed it and owned the decisions.
Faculty needed question sets that worked outside ReQuesta. I used an agent to organize rules for five formats, including QTI—the file Canvas imports.
How the structured source carries meaning
| Source convention | Carries |
|---|---|
| Headings | Document and question hierarchy, plus an accessible outline |
| Blockquotes | Quoted or supporting context, kept separate |
| Fenced code | Code, equation source, and preformatted text |
| Bold | A defined label or emphasis, with no font dependency |
| Lists | Question order and answer-choice order |
| Named labels | Metadata, answers, explanations, and other roles |
Markdown is the readable contract, not the whole data model. Developers built the field mappings for QTI and JSON and the exports for PDF, DOCX, and TXT.
Mappings built by developers I specified the rules; developers built the five exports. One source, five outfits.
The prototype showed where faculty needed control. The system helped that control survive implementation.
Gaurav Basantani
AI helped me organize the evidence. The final product judgment remained mine.
Teaching an AI agent to guide people, not overwhelm them.
Open to Product Designer roles anywhere in the US.
gaurav.inbox29@gmail.com