At PNHZ - Pavlodar Oil Refinery, I built a serverless execution engine connecting a Gemma-based annotation service with CVAT through REST APIs, reducing manual image-annotation work by 90%. I also deployed a Gemma VLM on a GPU cluster using Docker and vLLM.
I implemented a detection, classification and tracking pipeline that processed approximately 700 fps across 40 camera feeds on a single NVIDIA L40S GPU. I wrote automated tests, reviewed colleagues' code, and monitored the production system.
At Amsterdam UMC, I designed REST APIs and built a clinical NLP platform for real-time dialogue annotation, custom taxonomies and knowledge-graph visualisation. I also developed a long-form Dutch ASR pipeline and synthetic clinical-dialogue data pipelines.
My research has included coronary angiography analysis and clinical imaging datasets. I developed a vessel-analysis pipeline and fine-tuned PaliGemma 2, and helped develop MAGIC-CT with clinicians and researchers.
