Systems must work in practice.

For more than a decade at Liip, I worked on more than 25 successfully completed client projects. They include work for financial companies, major retailers, industrial businesses and tourism organisations.

My responsibilities ranged from software development and systems integration to cloud platforms, migrations, architecture, technical operations and consulting. This gave me practical insight into very different business models, technical starting points and decision-making structures.

For a revenue-critical commerce platform, I designed and implemented the Azure and Kubernetes infrastructure, deployment pipelines and scaling. The platform went live without service interruption. Since then, I have supported its operation and advised executive stakeholders on security and technical decisions.

For a tourism platform, I moved the application and infrastructure from an externally managed Azure account into one owned by the customer. This required understanding, changing and checking several hundred settings and connections between content, search and data systems. That change also happened without interruption. The technical migration then developed into advice on data architecture, cost, responsibility and AI readiness.

This breadth shapes my view of AI: a solution must fit everyday work, the existing systems landscape and the organisation’s responsibilities. Data access, operations, cost, failure, ownership and migration belong in the design from the beginning.

AI searches, compares, structures and prepares. Knowledge becomes durable only after deliberate review. External actions require approval. My own working system shows how these principles work together in daily use.

The work shows how model choice and access can remain separate from the knowledge architecture. It demonstrates architecture and prototype work.

I built and published a complete application with a Vue interface and Node.js backend. It prepares audio with FFmpeg, uses Azure Speech Services for transcription and speaker separation, and returns timestamps for individual words. Docker, GitHub Actions, Azure Container Registry and Azure Container Apps make deployment repeatable. The source and deployment setup are public.

In a separate open-source project using Whisper Large v3, I also added build and deployment support for Azure. The public fork containing my Azure work keeps that contribution distinct from my own application.

Together with three partners, I developed Personadeck from its first idea to a publicly available service. I initiated the work and principally drove product development. Within the team, I was responsible for all AI and prompt engineering work, as well as the backend, deployment, pricing and payment integration.

Personadeck turns a small set of inputs into detailed personas. I designed the AI workflows, prompts and quality rules that structure those inputs, generate results and prepare them for practical use.

Visit Personadeck

I worked on the YellowCube adapter and extensions that connect online shops with Swiss Post warehouse and shipping processes. The important part was the full flow: product data, inventory, orders, shipping and responses must remain consistent across several systems.

I explained the integration in the Liip article How YellowCube works. The open PHP package shows the technical foundation.

I spent eight years at UBS as an IT Business Analyst. I analysed banking processes, clarified requirements and translated them into practical specifications for software. Testing, documentation and support for existing systems were also part of my work.

This period is an important part of my professional foundation. I learned early that a technical solution only works when it respects the business purpose, the rules and the people involved in the workflow. Today, I apply the same thinking to architecture, cloud and AI.