I design AI people can trust.
Hi! I'm Marina :) Six years across consultancies, studios and in-house teams. Today solo designer on an enterprise AI platform, owning it from research through to spec.
- 6+ yearsdeveloping complex enterprise systems — consultancy, studios, in-house
- B2B SaaS
AI productsEnterprise AI platforms, search, and data-heavy workflows with operational weight - Solo designerI own the full lifecycle — research, specs, roadmap, design system — and prototype in code
- I teachthe advanced UX/UI course at ReDi School Munich, and own the material each year
AI is only useful when you can trust it, and you can trust it when you can trace it. So I start where the system is least sure: show what it is doing, name its confidence, give people a way to check it, and leave the decision with them.
Six years of this has taught me to design for systems that don't behave the same way twice.
Selected case studies
All case studiesMultilingual Detection & Filtering
AI Behaviour / Edge Case MappingEnterprise Search industryDesigning an AI-driven system to detect and filter content across languages, improving multilingual search relevance at enterprise scale.
Read the case
Sources: trusted context
AI Transparency / Retrieval UXEnterprise Search industrySeparating what the AI looked at from what it actually used, so users can judge an answer at a glance instead of re-verifying every document.
Read the case
Document Metadata Enrichment
Permissions / Designer + PMEnterprise Search industryFlat permissions let anyone create a label, so duplicate metadata multiplied on its own. I split it into three governed layers and removed the permission behind the pollution.
Read the case
Workspace Skills
AI workflows / Interaction designEnterprise searchReady-made actions — summarise, translate, compare — plus skills a client configures once, like filling in a questionnaire from their own documents. They run while the conversation keeps going.
Read the case
How I work
I work close to engineering, on products where the behaviour is the hard part. I use AI to explore far more options than I could draw by hand, and treat a shipped AI feature as the start of the design work rather than the end.
Learn how the system behaves
A model’s real behaviour shows up in production, across content nobody thought to test. I read that as the brief for the next design.
Vibe code every option I can think of
Instead of two or three hand-made proposals, I build five or more working options in code. Stakeholders react to something real, so we converge faster.
Then do the ground design work
Once a direction is chosen, I build it properly in Figma — components, states, and the design system it has to live in.
Write the rules with engineering
System behaviour is a design decision. What happens when the model is uncertain gets defined in the room with the people building it.