About
I spent twelve years asking "how do we know this is good enough to ship?"
I am Dávid Szemán, based in Budapest. Nine of my twelve years in quality were at HBO — HBO GO, then HBO MAX — where I went from being the sole in-house QA engineer for vendor-delivered releases to a QA Manager running international teams across EMEA and the US. The work was cross-platform: mobile, web, TV and console, covering onboarding, subscription billing, payment integrations and platform certification, on a service where a bad release is visible to millions within minutes.
After that I led test delivery at Manpower Hungary on confidential client engagements — first a 45-person team, then multi-lead governance across an organisation of 70+ testers. I designed the KPI framework, the risk-based prioritisation model and the release gates. The AI work started there, and nobody asked for it: I built an LLM reporting pipeline because senior leads were writing the same status report by hand every morning, and an AI-assisted framework for generating structured test content because volume was the bottleneck. Neither was in my job description.
My role ended in a collective redundancy in March 2026. I decided to use the time to go deep rather than take the first adjacent job: to work with these systems daily, across several vendors, on real problems rather than exercises — and to build the evaluation and cost discipline around them, which is the part that does not come from reading. Everything on the Work page came out of that period or the production work before it. What I want next is the upstream half of the job — framing the problem and designing the path through it, which in AI is where the expensive mistakes are made.
Honest gaps
What I am not
A portfolio that only lists strengths is a portfolio you cannot trust. These are the things I would raise myself in a first interview, each with what I would offer instead.
| Gap | What I bring instead |
|---|---|
| I have not run a company-wide AI transformation programme | I have run delivery organisations of up to 70+ people and owned release quality gates, KPI frameworks and vendor governance in regulated, multi-vendor environments. The stakeholder work transfers; I would be learning the scale of the AI programme, not the discipline of running one. |
| No model training or fine-tuning experience | My work sits above the model: choosing whether it belongs there, engineering the prompt and context for reliable structured output, and building the evaluation around it. If a role needs someone to train models, that is not me and I will say so in the first conversation. |
| No production vector database or RAG deployment at enterprise scale | I have built retrieval-shaped pipelines with real candidate sets, ranking and eligibility rules, and measured them with reference sets. The retrieval concepts are familiar; the enterprise-scale infrastructure is not something I have owned. |
| No formal computer-science degree (a Hungarian FOSZK vocational diploma in network and system administration) | Twelve years of quality and delivery leadership, and everything on this site was designed and built rather than described. I state the diploma level accurately rather than rounding it up. |
Client and product names from confidential engagements are deliberately absent from this site, as are internal figures. Where a case describes work done for an employer, it describes the practice, not the client.
Contact
Talk to me
Especially if you are deciding which AI use cases deserve funding, setting the bar a pilot has to clear, or trying to get something out of pilot purgatory.