AI is not a belief system. It is a changing engineering material.

Human2ai turns natural-language requests, photographs, PDFs and spreadsheets into structured product data, storefront changes, print-ready labels, product imagery and scheduled campaign assets while keeping consequential changes reviewable. I own the interaction model, shared component architecture, brand tokens, implementation patterns and release decisions across the product.

The core problem was organizational isolation: researchers could not see what other teams had already learned, so knowledge stayed buried and work was repeated. I designed the information architecture, shared taxonomy, metadata model, normalization rules, search and preview experience, tagging, subscriptions, governance and automation that connected those silos. This structure prepared twelve years of research for retrieval-augmented generation as well as human discovery, turning fragmented research into reusable enterprise evidence inside a regulated bank.

Domestic wire origination had to reconcile complex compliance, branch and system requirements while replacing a legacy blue-screen workflow with a modern, comprehensible employee experience. After validating the Axure prototype, I led the design transition while a UI designer created the high-fidelity Figma work. I also facilitated a micro-frontend workshop with developers to define reusable components, patterns and variants. Digital auditing capabilities and a digital-signature flow replaced printing and manual filing with a traceable record, eliminating approximately 1.56 million printed pages annually across 1,158 branches.

The prototype reimagined a form-heavy booking flow as a guided dialogue. It demonstrated how an AI assistant could gather trip requirements, maintain context, present options and support visual seat selection without losing the clarity and control customers expect from a transactional travel experience.

Co-creation workshops explored how AI, beacon technology and predictive analytics could deepen visitor engagement. I structured the ambition into an actionable path from immediate improvements to a personalized future-state experience.

I personally conducted all 64 moderated user-testing sessions across the flyer-status archetypes, interviewing each participant after the test, and designed the operational system around them: a searchable participant panel, repeatable recruitment, Mailchimp communications, Doodle scheduling, Axure prototypes, Webex sessions, recordings, early AI transcription through Otter.ai and searchable synthesis. The pipeline made continuous evidence gathering practical while shaping onboarding, points redemption, family pooling and seat-selection flows.

The North Star system map showed how data lakes, benefits platforms, IBM Curam, SADA and predictive tools could support a single citizen touchpoint. It gave service and technical teams a shared architecture for centralized data, proactive eligibility notifications and iterative modernization.

Matti tested how a LiveAvatar / HeyGen video agent, paired with ElevenLabs voice and conversational AI, could feel approachable while remaining bounded. The kiosk adapted language and complexity by class level, cleared private state between sessions and connected high-risk conversations to explicit, school-controlled adult handoff paths. A protected teacher UI helps staff respond appropriately to bullying and conflict by guiding the first response, documentation, escalation route and relevant school channels; it supports staff judgment rather than automating discipline.

Klara is a volunteer-built AI assistant for OMAS GEGEN RECHTS Hürth. It takes a slogan, quotation or political claim, recognizes the concern behind it, checks the factual basis and separates fact, analysis and conclusion. Three modes support responding to slogans, explaining political statements and creating teaching materials. A ten-level clarity scale adapts the same question from kindergarten through research level, while source transparency, explicit uncertainty and an empowering-not-preaching stance protect user judgment.

I built the workshop framework, implementation playbook, taxonomy and project hub, led more than 20 accessibility user interviews, then trained delivery leads to use the system. Research on phone anxiety also informed more inclusive, chat-first service pathways.