About ULTRA
User-friendly and trustworthy AI, tested in Flemish practice
ULTRA is a two-year VLAIO TETRA project (1 October 2025 to 30 September 2027) of Odisee, KU Leuven and Arteveldehogeschool, with a learning network of Flemish companies. This site is its pattern library and the first building block of the AI design guide: patterns with receipts, a compass for your context, and a growing set of cases.
This site is in English so that the three partners and the companies in the network share one vocabulary. Ask ultra@odisee.be for a Dutch summary of any page.
Who this is for
One site, three readers. Everyone sees the same cards. Each card has a second view, the research side, with the properties the pattern targets and the contexts it was seen in, and for six cards so far the mechanism and the references; the View switch on the card opens it.
Companies
Building or buying an AI feature
Start with Find my patterns, copy the words on screen and the component sketch, run the ten-minute metric card, and tell us what happened. Everything here except the logos may be used in your product: content CC BY 4.0, code MIT. Co-design sessions are part of the project for members of the learning network.
Researchers and designers
Looking for the mechanism
Every card has a research side: the properties it targets in the shared vocabulary and the contexts it was seen in; six cards so far also give the mechanism and the references. Switch the lens on any pattern.
Advisory group and VLAIO
Following the project
The practice cases, three of six slots filled so far, the partner studies, the candidates people vote for and the changelog show where the work stands. Every label on this site is derived from data, never typed.
Go deeper
Five pages that show the working parts of this site.
Find my patterns
How the compass decides
The two rule tables behind every recommendation, with the source of each rule.
How we test
Plan your test
Five steps to choose what to measure with your own users, and how many you need.
Research side
Vocabulary bridge
The KU Leuven property names next to the plain words on the cards.
Context
Libraries we build on
Six pattern libraries, what each is good for, and what this site adds.
Measure it
Nine metrics
Ten-minute kits with items, log events and benchmarks.
How a pattern earns its label
Four labels, one shape. The fill tells you how much evidence sits behind a card, and the receipt on the card tells you the rest.
A controlled comparison means two versions of the same screen, tested with the same people or with randomised groups, so that a difference can be attributed to the pattern and not to the people. A pattern is called Tested only when that comparison has a pre-specified primary metric, a complete study card, and more than sixteen participants. That floor comes from the 2025 systematic review behind the research side, which finds sixteen or fewer too small to test a hypothesis. Below that floor the same study is shown as Tested (pilot). The floor is a minimum, not a guarantee: how many participants a study needs depends on the effect you expect and on the design. This site avoids words that claim more certainty than a study gives.
How we test
In the UX Lab at Technologiecampus Gent or at the company. The same five steps for every case.
- Your case and questionA product with an AI feature, real users, and one thing you want to know.
- Two versionsThe pattern and its most common alternative, built as clickable screens.
- UsersTask scripts, think-aloud, interviews, interaction logs, eye-tracking where it helps.
- The study cardA pre-specified primary metric, the groups, the methods and the limitations, written before the analysis.
- The receiptWhat happened, per user group, on the pattern card and on the case page. Plan your own test.
Three partners, two families of patterns
Interaction patterns say who starts, what the user can override and how confidence and hand-offs appear. Explanation patterns say what the system shows to justify an output. Every card names a practice editor; a science editor from KU Leuven Augment is added once the research side has been reviewed.
Odisee, User Centered Experiences (UCE)
Project lead. Interaction patterns, the user tests in the UX Lab, the compass and this site.
KU Leuven, Augment
Explanation patterns, the taxonomy and the human-centred evaluation framework behind the research side of every card.
Arteveldehogeschool, Communication, Media and Design expertise network
Leads work package 3, the responsible use of AI in the design process: the report on AI in user research, the tool report, the selection wizard, the AI prototyping newsletter and Workshop 3. Designs the guide.
With the support of VLAIO, Flanders Innovation and Entrepreneurship. TETRA project HBC.2025.0124: TETRA projects translate research into practice for Flemish companies through a learning network and hands-on testing with companies.
Project pages: ULTRA project site · ULTRA at Odisee · ULTRA at Arteveldehogeschool
Timeline
Where the project stands and what arrives next on this site.
- October 2025Project start. Three partners, a learning network of Flemish companies, two years.
- Autumn 2025VAIA FormFlow study in the Odisee UX Lab with eye-tracking and think-aloud; the poster was presented at ACM IUI 2026 in March 2026.
- 2 June 2026Workshop 1, Discovery and strategy, with the learning network.
- Summer 2026CNEXT user tests with accountants on AI that turns e-mail into tasks; FORCIT case on helping people describe a challenge for expert matching; Goose partner study.
- September 2026A preview of this site for the partners.
- 15 October 2026Partner meeting with the advisory group, followed by the second meetup, Designing & Evaluating AI Experiences, with In The Pocket at Technologiecampus Gent.
- 23 October 2026Workshop 2, Design: smart AI interface patterns that build trust. Ultra Patterns goes public the same day.
- January 2027The human-centred evaluation framework of KU Leuven Augment becomes the research side of every card.
- February 2027Workshop 3, Build, led by Arteveldehogeschool. Date to be confirmed.
- 20 May 2027Workshop 4, Validate.
- Summer 2027The AI design guide, online and in print, built from these pages; more cases and kits.
- 30 September 2027The project ends.
How this site maps to the project
For the advisory group and VLAIO: where each part of ULTRA lands here.
| Project part | Lead | On this site |
|---|---|---|
| WP1 · Taxonomy and human-centred evaluation framework | KU Leuven Augment | The research side of every card, the vocabulary bridge and the metric cards. The full framework is published in January 2027. |
| WP2 · Design guidelines and the design framework | Odisee UCE | This site: the patterns, the receipts, the compass and the cases. |
| WP3 · Responsible use of AI in the design process | Arteveldehogeschool | The report on AI in user research (January 2027), the tool report (March 2027) and the AI-tool selection wizard (May 2027) link here when they are published. Arteveldehogeschool also leads Workshop 3 and writes the AI prototyping newsletter for members. |
| WP4 · Practice validations, the six practice cases | Odisee UCE with the companies | The six practice cases and their receipts, tracked on the cases page. |
| WP5 · Knowledge sharing and the AI design guide | All three partners | The four workshops in the timeline. The AI design guide (summer 2027) builds on this site. |
Bring your case
Bring the next receipt
If your product has an AI feature and real users, send three screenshots and one sentence on what you want to know. Open slots remain in the six practice cases, and co-design sessions are part of the project for members of the learning network.
The address is ultra@odisee.be. Not yet in the learning network? Write to find out how companies join. Ask to be added to the ULTRA newsletter; members also receive the AI prototyping newsletter of Arteveldehogeschool.
What to expect. Nothing about your case is published before you approve it, screenshots stay with us until then, and your users' data never appears on the site. Co-design sessions are part of the project for members of the learning network.
Changelog
What changed on this site, newest first.
- 2026-10-29 · Ultra Patterns moves to patterns.ultra-project.be, next to the ULTRA project site. The full list of patterns is now at /library and every pattern has a short address such as /match-recipe. Old links still work.
- 2026-10-23 · Version 1: Ultra Patterns goes public on the day of Workshop 2, with 19 patterns, nine metric cards based on the KU Leuven evaluation framework, the compass, the VAIA case and the Goose partner study.
- 2026-09-08 · CNEXT Mail2Flow: the July usability test as a full case page (method, task completion, ten findings, the report's recommendations), three new patterns (17 to 19) with a pilot receipt, and five new candidates. Case pages now carry structured sections; unapproved cases show on preview builds only.
- 2026-09-03 · Goose partner study: six conversation patterns, a partner case and three candidates. Evidence legend, partner logos, redesigned compass and about pages.
- 2026-09-02 · Preview: ten patterns, nine metric cards, the compass with two answers, the VAIA case.
Research
Van den Bossche, T. and Abdul Khalil, A. (2026). From Pattern Libraries to Practice: Context-Aware Selection of AI Interface Patterns. IUI Companion. DOI.
Donoso-Guzmán, I., Ooge, J., Parra, D., Verbert, K. (2023). Towards a Comprehensive Human-Centred Evaluation Framework for Explainable AI. arXiv.
Donoso-Guzmán, I. et al. (2025). A Systematic Review of User-Centred Evaluation of Explainable AI in Healthcare. arXiv.
Amershi, S. et al. (2019). Guidelines for Human-AI Interaction. CHI 2019.
Parasuraman, R. and Manzey, D. H. (2010). Complacency and bias in human use of automation: an attentional integration. Human Factors, 52(3).
Tversky, A. and Kahneman, D. (1974). Judgment under uncertainty: heuristics and biases. Science, 185(4157).