Agentic LLM systems, and computer vision for oncology, genomics, and healthcare, grounded in trust and transparency.
Since July 2025 I’ve been Co-CEO of MediBit, a five-person team of engineers, a postdoctoral cancer researcher and an oncology resident. We build AI software for healthcare, from clinical-trial matching in oncology to computer vision for ENT telemedicine.
Matches structured and unstructured oncology patient profiles to clinical-trial eligibility criteria, so hospitals, CROs and sponsors can find eligible patients faster.
My role · product vision, pilot strategy, hospital and partner negotiations, roadmap from prototype to deploymentAn AI throat-scanning application for ENT telemedicine, built with Larinta (formerly Shoelessly). Computer vision analyses throat images to support ENT digital diagnostics.
My role · project manager: scope, timeline, client communication, team coordinationSpecialised agents, knowledge graphs and RAG that retrieve and synthesise evidence from literature, variant databases and guidelines, with every step reviewable by an expert.
MSc thesis · 2025 →My deepest experience is in digital pathology: segmentation and quantification in H&E and IHC images, often with weak labels. I’m interested in medical imaging across modalities and applications, which now includes throat images for ENT telemedicine at MediBit.
BSc thesis · EHH IKEM · MediBit × LarintaTrial matching and patient-handover tools built for clinical governance: verbatim source quotes, deterministic checks, human confirmation, audit trails.
MediBit · Rakathon · 2025 →Long before I wrote code for hospitals, my first research project was on why some testicular cancers resist chemotherapy. Every step since has come back to oncology from a different angle.
F. Duvač, P. Farkaš, D.-D. Horvath, D. Ivan, A. Komanek and M. Sivák. Proc. 2026 IIT.SRC, Bratislava, Apr. 2026. IEEE.
doi:10.1109/IIT.SRC67760.2026.11628845Comfortable across the whole path from research to product: data, model, agent workflow, front end, and the stakeholder conversations that get it deployed.