I help product teams take LLM features from “works in the demo” to something they can run in production: reliable when the provider misbehaves, affordable as usage grows, and safe enough to put in front of users.
Behind that sit 25+ years of building production software, the last 18 or so as an architect and CTO. My core stack is TypeScript and Node across front and back end, with PostgreSQL and queue-based systems underneath. I work as a freelance architect from Vienna, with clients in Europe and the US.
How I work
I don’t sell short projects. Clients bring me in for the long run, and my engagements have lasted two to five years and more.
-
It starts with one concrete project
Usually something a team can’t staff with the seniority it needs: a migration, a new system, an AI feature that has to reach production.
-
I deliver it from inside your team
Remotely from Vienna, in your codebase and your workflow, working in English across time zones. I own the architecture and see it through to production rather than handing over a report.
-
The role grows with the trust
Once the first project lands, teams start bringing me their other architectural decisions. Over time I become the senior architect they keep, across initiatives.
That continuity is the point. You get someone who knows your system, your people and your history, without hiring a full-time architect. I keep the number of clients small so each one gets that attention.
What that looks like today
Forbes · 2021 – present
I was brought in as a senior engineer to lead the front end’s move to React and Next.js, and was promoted to Senior Software Architect in 2023. Engineering and product teams now bring me their architectural decisions, and I’m involved in every running initiative. That includes architectural oversight of AI-heavy systems built by a separate team: reviewing designs, weighing trade-offs and watching how they behave in production.
Liza.network · 2018 – 2021, 2024 – present
I was CTO of the platform from 2018 to 2021. In 2024 they brought me back to architect its AI layer, which I designed and built and which is now live in production.
- A two-stage matching pipeline: vector search narrows each new entry to a small set of complementary candidates, and an LLM evaluates only those. AI cost stays bounded as the platform grows.
- Embeddings kept in PostgreSQL, so the same vectors power hybrid semantic and full-text search with no extra service to run.
- A provider-agnostic AI layer with asynchronous processing and per-call monitoring of latency, errors and token usage.
A detailed case study is available on request.