MaterialSpace · Master’s thesis · Human-AI Interaction

How can AI be designed for non-experts while maintaining usability, trust, and acceptance?

I led the research, product design and development of MaterialSpace across four studies with 60 participants. The final design replaced open prompting with configuration, clarification and curriculum-grounded generation, reducing the task from repeated prompt iterations to one guided flow.

Human-AI InteractionTrust in AIProduct DesignDevelopment

RoleResearch, product design, development · solo

ContextMSc Human-Computer Interaction · PLUS and FH Salzburg

Scope4 studies · 60 participants · Germany and Austria

ResultGraded 1.0 · merit scholarship

LinkThesis · University of Salzburg ↗

Impact

60ParticipantsThe concept was researched and validated across four studies, including a survey, interviews, a pilot and a final evaluation in Germany and Austria.
2–4 → 1AttemptsRepeated prompt iterations were replaced by one structured flow in which intent was clarified before generation.
8 of 8Trusted it in useIn the final evaluation, all eight participants reported feeling safe and saw no reason for caution when interacting with the system.
75%Wanted to keep itAfter using MaterialSpace, three out of four participants indicated that they wanted to continue using the product.

Overview

Context

An AI worksheet tool for lower secondary teachers, built and evaluated end to end.

A master’s thesis in Human-Computer Interaction. The question: can an artifact-centered assistant hold usability, trust and acceptance at once?

ChallengeHow I got past it

Non-experts cannot phrase promptsConfiguration replaces the prompt
Output cannot be judgedThe curriculum grounds every result
15 minutes or 40, same taskIntent clarified before anything generates
Trust decides adoptionFive requirements, validated with 60 people

Solution

Structure instead of prompting.

Trust decides whether an AI product gets used.

01 · Introduction & relevance

Product Strategy

88 percent already use AI. Two in three name rework as their biggest barrier.

Classrooms in Germany and Austria are going digital, and worksheets are turning into interactive, multimodal documents. I surveyed 42 lower secondary teachers on how they build them.

Almost all of them already work with AI. The dominant interface is a chatbot, which asks a teacher to express a whole lesson through one text prompt. What comes back often misses the topic or the grade level.

Challenges with AI use

The four biggest barriers, and what answered each

Post editing needed67.6 %Editing tools per section instead of regeneration
Too little time48.6 %One pass: configure, clarify, generate
Topic did not match43.2 %Curriculum documents as context, not as prompt text
Unclear how to apply27.0 %Clarifying questions name the curriculum requirement

Seven further barriers were each named by under 25 percent. n = 37 teachers.

So the tool creates a second task instead of removing the first. That rework is the cost this thesis set out to remove.

UCD process · 4 studies, 60 participants

One user-centered cycle, end to end

Pick a phase.

Named the loop: write a prompt, read the output, rewrite it.

The persistent worksheet canvas, built rather than sketched.

The real cost: fifteen minutes or forty, depending on the prompt.

MaterialSpace is my answer: an artifact-centered assistant, built and evaluated with 60 participants across four studies. The bar I set for it was one usable, curriculum-aligned worksheet in the first session, without any prompt training.

Process

Six phases, 60 participants, three builds

Above the line I built. Below the line I learned. The pilot forced two more builds.

the two builds the pilot forced

Phase 1Exploratory ResearchN = 42 survey · N = 4 interviews

Phase 2Design & Implementation IPrototype v1

Phase 3Test & IterationN = 6 pilot study

Phase 4Design & Implementation IIclarification dialog added

Phase 5Design & Implementation IIIper-section editing tools

Phase 6Final EvaluationN = 8 · Trust Scale, TAM

↑ built · ↓ learned42 survey · 4 interviews · 6 pilot · 8 final evaluation

02 · Process

User Research (Qual + Quant)Usability Testing

The pilot found the real cost. Fifteen minutes or forty, depending on the prompt.

I surveyed 42 teachers, interviewed four and compared eleven tools. All three pointed at the same loop: write a prompt, read the output, rewrite it.

The first prototype broke that loop. A worksheet canvas that stays on screen, context set once, every result validated against a schema.

Lo-fi wireframes · the model decision

Chat or worksheet canvas, decided at wireframe stage

Flip between them.

Then the pilot. Six teachers, one task: a percentages worksheet for grade 6, with a pencil sketch as context. The dropdowns worked for all six. The free-text field did not.

P5 typed four lines with a worked example and finished in 15 minutes, first attempt. P4 typed “Task seven. Edit.” and needed four attempts and 40 minutes. Same task, same interface.

The way to a worksheet

40 minutes and four tries, or 5 and one

Press play or drag.

0 min

BeforeContext in the prompt · 2 to 4 attempts

PromptResultReworkPromptResultRework

↰ back to the prompt↰ back to the prompt↰ back to the prompt

Worksheet after 40 minutes and 4 attempts

AfterContext in the interface

ConfigureClarifyWorksheet

Worksheet after 5 minutes

02040 min

“Let’s hope it understands what I want.”
P3 · mid-task, pilot study

Spread · pilot, N = 6, same task

Same task, same interface. The prompt decided.

the average · it would hide the spread

P5 · four lines with a worked example1 attempt

15 min

P4 · “Task seven. Edit.”4 attempts

30–40 min

010203040 min

Only prompting skill separates the two rows.

03 · Design Requirements

Interaction DesignHuman-Centered AI

Five requirements that replace prompting with structure.

Design requirements

Five requirements that replace prompting with structure

The straight path was to generate on request and let teachers edit afterwards. That was the pilot version, and it cost two to four attempts per task. DR5 replaced it.

Clarify first

Nothing is generated before the intent is named

Request

topic, subject, grade, difficulty, curriculum PDFs

Clarifying questionsthe step I added

01Grammatical focus · much, many or the contrastcites the curriculum
02Thematic context · the drawn bottle and the applescites the curriculum
03Task types · which interactive formatscites the curriculum

1 to 6 questions, then a proposed structure of at most 3 to 5 sections

Worksheet

schema-validated, section by section

Try it outDrag a word into a column, or tap a word and then a column.

Lehrplan *Bayern⌄

Schultyp *Mittelschule (R)⌄

Fach *Englisch⌄

Jahrgangsstufe *7⌄

Thema *Mengenangaben: much, many⌄

SchwierigkeitAnfänger

Context image · a milk bottle and three apples

Sort the words from the picture into the correct column:

Countable

Uncountable

Inhalt

01Einstieg: Zählbar oder nicht?
02Bildfragen
03Merkregel: much oder many?
04Wortschatz-Check: Countable or Uncountable?
05Übung: Much or Many?
06Anwendung: Ask Questions!
07Quick Check

1Pick a word up. The words, the columns and the solution arrive together, as data, or not at all.
2Drop it in a column. The marking is free: the answer came with the task.

04 · Results

Analytics & UX Measurement

8 of 8 trusted it. 75 percent wanted to keep it.

Results

What the design measurably changed

0 of 8trusted it in useboth risk items of the trust scale · the best possible score
0%wanted to keep using itafter the first session
2–4 → 1attempts per worksheetguided clarification instead of exploratory prompts
0prompts writtenthe whole context lives in the interface

Human-Computer Trust Scale · N = 8

Every single teacher felt safe

Felt safe interacting with it8 of 8
Saw no reason for caution8 of 8
Expected no negative consequences7 of 8

Strongly disagree
Disagree
Somewhat disagree
Neutral
Somewhat agree
Agree
Strongly agree

Negative consequences (R)
Must be cautious (R)
Risky to interact (R)
Acts in my interest
Help when needed
Understands my needs
Competent and effective
Fulfils its support role
Has all the features
Can fully rely on it
Reliable support
Trust the information

Items marked (R) were reverse-scored. Shares of N = 8.

Try it outChoose much or many in each gap, then press Check.

Lehrplan *Bayern⌄

Schultyp *Mittelschule (R)⌄

Fach *Englisch⌄

Jahrgangsstufe *7⌄

Thema *Mengenangaben: much, many⌄

SchwierigkeitAnfänger

Fill in: much or many?

Context image · BAKING TIME

A: “Let’s bake a cake! How apples do we have?”

B: “Not . Only two. We need more.”

A: “Okay. And how sugar is left?”

B: “Are there bags of flour left?”

A: “No. And there isn’t milk, either.”

Inhalt

01Einstieg: Zählbar oder nicht?
02Bildfragen
03Merkregel: much oder many?
04Wortschatz-Check: Countable or Uncountable?
05Übung: Much or Many?
06Anwendung: Ask Questions!
07Quick Check