UI elements of the portal on an orange background: illustrations, career-changer guide, radius search, qualification entry and entry buttons
Built with anAI-based Workflow

A portal to tackle the skills shortage:
400+ training programmes in child & youth welfare

My role: UX design and frontend, from research to the launch campaign.

The Karriere Kinder- und Jugendhilfe portal on laptop and smartphone
50 %+

of consultation appointments were booked within the first 2 weeks

17 % → 80 %

of test participants knew where to start after the redesign

3K

visitors in the first 4 weeks after launch

Facts

Development project 2025 to 2027
HAWK, Faculty of Social Work (ZZHH)
Funded by the Lower Saxony Ministry for Social Affairs, Labour, Health and Equality

Team:
2 project leads, advisory service, development, UX design

Methods

Personas · customer journeys · reconstructing consultations · prototype as a basis for discussion · information architecture · wireframing · prototyping · feedback forms · moderated usability tests (think aloud) · unmoderated online test · affinity mapping · severity prioritisation · How Might We

Tools

Figma · FigJam · Claude Code · Claude Cowork · Figma Weave · Adobe Firefly · NotebookLM · Illustrator · Premiere Pro · GitHub · Jira

Problem

Career changers have no clear way in.

Skills shortage

Providers can’t find qualified staff for many of their vacancies; career changers are urgently needed.

Legal hurdles

Regulations make it hard to get in; even providers often don’t know who they’re allowed to hire.

400+

Scattered knowledge

Anyone looking for information has to click through lots of websites or know someone who knows the field.

First approach

What AI suggested

An AI chatbot seemed the obvious choice …

The content was already there, so in theory a chatbot could simply have drawn on it. That’s why in our first user flows it was a fixed component, and at the sector conference providers asked for a chat too.

# Prompt

What would you build into the portal so that people get personalised answers about starting a career in child and youth welfare?

# One of the first outcomes

AI-powered advice assistant

A chat that answers individual questions about further training.

Excerpt from the first user flow (in German): after “field of work selected” comes an “interview/chatbot” step that lists the qualification requirements. The question “Entry requirements met?” then branches into yes, no and not sure, with chatbot replies such as “Then let’s look together at what options you have.” If no, it asks whether the person is interested in further training.
From the first user flow

Insight

After testing, interviews and user flows

No to adding yet another AI chatbot, because …

Hallucinations in testing

When information was missing, the test chatbot made it up, despite clear instructions. Even after talking to IT, there is no way to rule this out (yet).

Worst case: the wrong course

If the chatbot gives the wrong answer to “Can I work here with this qualification?”, someone might start a training programme that doesn’t legally qualify them for the job they want.

People wanted personal advice

For a decision like this, the target group would rather talk to a person. The capacity is there, and a new portal has to earn trust first.

Solution

Filterable overview
and personal consultation

That’s why I chose a filterable overview, with filters for entry requirements, work experience and interests. Where that isn’t enough, users go straight to a personal consultation.

  • 50 %+ of consultation appointments were already booked within the first 2 weeks

AI-based advice isn’t off the table. The foundation comes first; later, AI can help at specific points.

Mobile qualification overview in the portal: search field, active filters “degree” and “26203 Wardenburg · 40 km”, 38 of 106 entries, list, map and saved-list views, below them degree programmes such as Special Education B.A. in Oldenburg

User test

A good overview only helps if people can find their way in.

  • Round 1 · moderated user test 1 of 6 knew where to start
  • Round 2 · moderated user test 4 of 5 started with the newly added quick-select buttons

Iteration

Between test rounds

Users should be able to find their bearings right away

What we tested

A home page with intro text and direct access to the content.

What failed

5 of 6 participants didn’t know where to start.

What we changed

Instead of explanatory text, the home page now asks “What are you looking for?” and offers quick-select buttons for career changers and providers.

Slide from the usability analysis: cluster “entry and orientation” with severity 4, 5 of 6 people affected, insight, observations, How Might We question and recommendations such as quick-select buttons
Before (low-fi)

Before and after the user test

After (high-fi)
After: portal home page with quick-select buttons at the top and below them “What are you looking for?” with two entry points
Before: prototype home page with heading, search field and an introductory text without a clear call to action

Result

From 4 spreadsheets and 400+ scattered training programmes
to 1 structured information portal.

Easy to grasp, clearly structured and logically built up.

Feedback from the sector conference

Workflow

Who did what in each phase
Phase Us AI
Research
  • Talking to the target group
  • Customer journeys
  • Reconstructing consultations
  • Created a fillable persona PDF
  • Chatbot testing
Structure and prototype
  • Building an information architecture
  • Structuring the portal for handover to the AI agent
  • Ideation with ChatGPT and Google Stitch AI
  • Prototype with Figma Make, Cursor and Claude Code
Usability tests
  • Moderated think-aloud tests
  • Unmoderated online test
  • Interview guide and Google Forms survey prepared with Claude Cowork
Analysis
  • Affinity mapping
  • Defined priorities and How Might We questions
  • NotebookLM summarises
  • Claude Cowork clusters and shows connections
Build and launch
  • Final decisions on design, content and accessibility
  • Planned the campaign and approved content
  • Built the site with Claude Code via GitHub
  • Checked accessibility and SEO
  • Images with Adobe Firefly and Figma Weave
  • Co-wrote site and campaign copy
  • Set up our own analytics

Reflection

Next time, start with
the spreadsheets

I’d go through the data carefully right at the start: does the structure work, does anyone actually need this information, or does it need to be described differently?

I’d have spotted many of the issues from the user test much sooner, along with the gaps that only became visible while building the overview.

Next project

Design System

1 design system, 9 themes: from campus project to university standard

One component set for the whole university, from the virtual campus to seven faculties.