For 30 years every website has worked the same way - unitl now.
For thirty years, every website has worked the same way: menus at the top, breadcrumbs to orient you, hierarchies to navigate. We have internalised this so completely it no longer feels like a design choice - it feels like the nature of the web itself. This project removes all of it.
It builds the first LLM-powered conversation-based university website. Prospective students simply ask what they want to know - "How do students work in class?" "Which projects explore social themes?"- and receive answers from actual programme content with browsable visual material embedded. The organising principle: meet people where their questions are, not where institutional structures place information.
This is not a chatbot widget. Conversation and visual browsing coexist as equal methods across every page, without menus, hamburger buttons, breadcrumbs, or search bars. Visitors scroll through portfolios, then ask contextual questions. The system builds memory across sessions. Research confirms no website in any domain operates this way.
Early navigation-less experiments used rule-based chatbots that matched keywords and followed scripts. Modern large language models changed everything: they understand semantic meaning, synthesise responses, adapt to nuance. This builds the first institutional website on that foundation.
The architectural challenge: when conversation constructs navigation paths dynamically, how do you preserve spatial context? Users click project cards from conversation results, explore detail pages, then return - expecting their conversation exactly where they left. This requires state management architecture that bridges two fundamentally different interaction paradigms without privileging either.
Technical implementation uses WordPress backend with Mistral API (GDPR-compliant, EU-based), sessionStorage for conversation persistence across page navigation, RAG pulling from actual programme content to prevent hallucination, and intelligent pagination that feels conversational rather than algorithmic (no "showing results 6–12 of 15"). The system auto-detects language (German/English), maintains conversation context when users navigate away and return, and handles edge cases where traditional and conversational modes intersect.
This positions at the frontier of what interaction designers now call "generative UI" - interfaces that construct themselves dynamically based on user intent rather than predetermined information architecture. But unlike current generative UI work focused on task-specific displays, this operates across a comprehensive institutional information space: course descriptions, student portfolios, application requirements, 200+ projects, faculty profiles.
The talk shares: concrete architecture decisions behind dual-mode coexistence, state management patterns for conversation persistence across navigation, pagination design that maintains conversational flow, and insights from development through initial deployment. For UX practitioners wondering what institutional communication looks like when AI isn't bolted onto existing patterns but fundamentally reimagines how people access complex information - this is a working answer grounded in comprehensive technical documentation and real-world implementation.
For thirty years, every website has worked the same way: menus at the top, breadcrumbs to orient you, hierarchies to navigate. We have internalised this so completely it no longer feels like a design choice - it feels like the nature of the web itself. This project removes all of it.
It builds the first LLM-powered conversation-based university website. Prospective students simply ask what they want to know - "How do students work in class?" "Which projects explore social themes?"- and receive answers from actual programme content with browsable visual material embedded. The organising principle: meet people where their questions are, not where institutional structures place information.
This is not a chatbot widget. Conversation and visual browsing coexist as equal methods across every page, without menus, hamburger buttons, breadcrumbs, or search bars. Visitors scroll through portfolios, then ask contextual questions. The system builds memory across sessions. Research confirms no website in any domain operates this way.
Early navigation-less experiments used rule-based chatbots that matched keywords and followed scripts. Modern large language models changed everything: they understand semantic meaning, synthesise responses, adapt to nuance. This builds the first institutional website on that foundation.
The architectural challenge: when conversation constructs navigation paths dynamically, how do you preserve spatial context? Users click project cards from conversation results, explore detail pages, then return - expecting their conversation exactly where they left. This requires state management architecture that bridges two fundamentally different interaction paradigms without privileging either.
Technical implementation uses WordPress backend with Mistral API (GDPR-compliant, EU-based), sessionStorage for conversation persistence across page navigation, RAG pulling from actual programme content to prevent hallucination, and intelligent pagination that feels conversational rather than algorithmic (no "showing results 6–12 of 15"). The system auto-detects language (German/English), maintains conversation context when users navigate away and return, and handles edge cases where traditional and conversational modes intersect.
This positions at the frontier of what interaction designers now call "generative UI" - interfaces that construct themselves dynamically based on user intent rather than predetermined information architecture. But unlike current generative UI work focused on task-specific displays, this operates across a comprehensive institutional information space: course descriptions, student portfolios, application requirements, 200+ projects, faculty profiles.
The talk shares: concrete architecture decisions behind dual-mode coexistence, state management patterns for conversation persistence across navigation, pagination design that maintains conversational flow, and insights from development through initial deployment. For UX practitioners wondering what institutional communication looks like when AI isn't bolted onto existing patterns but fundamentally reimagines how people access complex information - this is a working answer grounded in comprehensive technical documentation and real-world implementation.
Got some juicy gossip?
We’d love to hear it (and any other questions, wishes or suggestions you have).
