Nineteen years, from assembler and bare-metal C to generative AI. Every few years the stack changed underneath me. What did not change: I learned the layer below the new thing until I could build on it with confidence.
Four years of automatic control systems, ending with a diploma project built from bare copper up.
BSc Informatics, 180 ECTS, with a thesis on image processing built as a working application rather than a paper.
MSc Informatics, study programme Software Engineering, with a thesis on an unconventional approach to image compression reaching roughly 400:1.
Nineteen years in two halves that rarely meet in one person. Ten years close to the metal — printer firmware in C across three processor architectures, then the GNU compiler and debugger themselves. Nine years upward through the stack — automotive code generation, Angular, JVM internals and garbage collection, distributed platforms on Quarkus and Kubernetes, and finally generative AI running in a product used by business travellers every day. The thread is the same throughout: build the thing underneath, and understand the constraints below that.
Ten years of low-level C, real-time systems and toolchains.
A summer internship spent writing a multithreaded backup server in C.
Two years building printer-side resource handling in low-level C for high-end production printers.
Porting IPDS functionality across x86, MIPS and PowerPC, on three different operating systems.
Four years on colour printing and colour management — the hardest correctness problem on the product.
Compiler and debugger development on the GNU toolchain for three embedded architectures.
Took on Eclipse CDT toolchain integration in Java while still writing compiler C — and started leading the team doing it.
Owned quarterly GNU toolchain releases, and automated the delivery process that made them repeatable.
Twenty-two months out of C — Java tooling, then a new country, a new industry and the only stretch of the career spent in the browser.
Eclipse-based Java tooling for AUTOSAR configuration, including R&D on the Adaptive AUTOSAR toolchain.
Moved from Romania to France and from embedded C to Angular, on a real-time airline merchandising platform.
Five months on Cytric Expense — proposed and introduced privacy-respecting web analytics.
Back into the JVM and out again into distributed systems — four years spent on how things are built rather than what they do.
Nearly two years on JVM performance, garbage-collector studies and production incident forensics.
Built the first version of the Cytric Travel integration for Microsoft Teams from a blank repository.
Rebuilt the Teams integration onto Amadeus' internal platform — the standardised route rather than a bespoke one — and shipped its first AI chatbot.
Technical lead for the Cytric AI Assistant and Epic Owner at programme level for the epics behind it.
A programme-level epic spanning three teams, isolating the Bot Framework behind a dedicated connector.
Modernised travel-policy question answering onto Azure AI Search and integrated it into the main product.
Prototyped an Agent Playground for local development and a personalisation microservice, before teams committed.
Sets AI architectural direction across the programme and consults for teams building their own AI capabilities.
Themes I keep coming back to, and want to write about properly. There will be more.
Every job I have taken has been one layer below the last one that made me curious. The board, the firmware, the compiler, the runtime, the platform.
A colour that is subtly off. A pause that only appears under real traffic. An answer that is fluent and wrong. The failures I have spent most of my life on are the quiet ones.
Compilers, generators, release automation, playgrounds for other people's experiments. Given the choice, I would rather own the tool than borrow it.
Older than it looks: a master's thesis that threw the pixels away and let a search find its own representation, and a day job doing a version of the same thing sixteen years later.
A personal AI assistant that knows my work and background — ask it about my experience, or leave a message and it will pass it along.