# Flavius Marc, Software & AI Engineer

Software engineer in Utrecht with over 7 years of experience at a global automotive supplier, a software consultancy, medical-technology startups and university hospitals: from embedded C/C++ firmware to medical device software and full-stack AI applications with LLMs, LangGraph, FastAPI and Supabase. Most recent work: a personal agentic hub with Hermes on a VPS.

- Full name: Flavius Gabriel Marc
- Location: Utrecht, Netherlands. EU citizen.
- Website: https://flaviusmarc.com/
- GitHub: https://github.com/flavius25
- Contact: https://flaviusmarc.com/#contact

## Roles I fit

### Software Engineer

Over 7 years of production software in C/C++, C# and Python. I started in automotive, writing software for radar and camera systems at Continental. Since then I have built REST services for payment terminals, contributed algorithms to cardiac imaging applications, and contributed to the software of a medical lab device at Xyall.

- Close to the hardware: Bare-metal firmware on multi-core ARM Cortex-A53, debugged with Lauterbach.
- Secure services: A C# REST API for point-of-sale transactions over TLS 1.3.
- Measured improvement: About 20% better barcode detection in the Xyall Tissector, through tuning and validation.

### AI Engineer

I build agentic systems that are meant to run in production. My most recent one is an AI learning platform on Google Cloud, where I did the orchestration in LangGraph, the tracing in LangSmith and the database work myself. Before large language models I trained deep learning models on medical images, and one of them contributed to a published paper. Over 7 years of software engineering sit underneath both. I also build the AI harness a project needs, fitted to its environment and its team: controllable agentic loops with feedback built in, or systems that learn from their own results and take on more of the work by themselves. The aim is always the same: faster development, a better understanding of the system and more autonomous work.

- An agentic product, end to end: Course generation and an AI tutor, orchestrated with LangGraph and deployed on Google Cloud.
- Deep learning that reached a journal: A segmentation model for cerebral angiography, published in IJCARS in 2025.
- Production habits first: Over 7 years of C/C++, C# and Python in automotive, fintech and medical devices.

### Forward Deployed Engineer

At Xyall I work on the software between a lab machine and the clinician's screen, and I reworked the installation process so that setting up at a customer takes less time and goes wrong less often. I have worked in a large corporation, a consultancy, a startup and two university hospitals, so I settle quickly into a new team and its constraints. I can also build the AI part myself, from the first prototype to a deployed system. I also build the AI harness a project needs, fitted to its environment and its team: controllable agentic loops with feedback built in, or systems that learn from their own results and take on more of the work by themselves. The aim is always the same: faster development, a better understanding of the system and more autonomous work.

- Shorter installs at the customer: Streamlined the Xyall installation workflow, cutting on-site setup time and making rollouts more reliable.
- Four kinds of organisation: Corporation, consultancy, startup and academic hospital, each with different users and rules.
- Builds the whole solution: Delivered an AI learning platform alone, from orchestration to database to cloud deployment.

## Experience

### Software Engineer, Xyall

Jun 2025 to now, Eindhoven, Netherlands. Medical technology: the Tissector automates tissue dissection from pathology slides for molecular diagnostics.

- Improved barcode detection for slide identification in the Tissector by about 20% by tuning and validating the detection algorithms.
- Built features and security improvements in the Flask broker service that connects the machine to the clinician-facing front end over REST.
- Streamlined the customer installation workflow, cutting on-site setup time and making rollouts more reliable.
- Develop the core machine software in C# and Python, with a focus on robustness and maintainability.

Worked with: C#, Python, Flask, REST APIs, computer vision, Cursor, Claude Code

More: https://flaviusmarc.com/experience/xyall.html

### Software Engineer (consultancy), Bitwise Europe

Dec 2023 to Apr 2025, Eindhoven, Netherlands. Software and test engineering for clients in fintech and assistive technology.

- Built a C# REST API for secure point-of-sale payment transactions over TLS 1.3.
- Extended and optimised a C++ Windows communication server that drives ticketing devices for testing.
- Recreated a client's development environment on Linux virtual machines to run and debug their software package.
- Designed test setups for a prototype sensor in an assistive walking device, collected the data and analysed it in Python to evaluate usability.

Worked with: C#, C++, Python, Java, Kotlin, Linux, REST APIs

More: https://flaviusmarc.com/experience/bitwise-europe.html

### Software Engineer (part-time), CART-Tech

May 2022 to May 2023, Utrecht, Netherlands. Medical imaging startup, cardiac imaging software.

- Worked on image-processing algorithms in C++ in a Qt desktop application for cardiac imaging: new features and debugging.
- Developed image-processing algorithms in Python from MATLAB prototypes, optimised for speed and scale, including 3D mesh generation and reconstruction.
- Worked directly with the technical manager so the application met clinical accuracy and usability requirements.

Worked with: C++, Qt, Python, SciPy, MATLAB, GitLab

More: https://flaviusmarc.com/experience/cart-tech.html

### Embedded Software Developer, Continental

Jul 2018 to Aug 2021, Timișoara, Romania. Automotive radar and camera systems (ADAS).

- Developed bare-metal firmware for radar and camera systems on multi-core ARM Cortex-A53, and debugged multi-core execution with Lauterbach TRACE32 (bootloader, symbols, inter-core coordination).
- Implemented and unit-tested C/C++ modules from UML detailed designs, with requirements traceability to automotive safety standards.
- Took part in code-quality analysis and design reviews with international, multidisciplinary teams.
- Spent my first month in system testing: black-box functional tests of an instrument cluster with CANoe.

Worked with: C, C++, ARM Cortex-A53, Lauterbach TRACE32, CANoe, CAN, UML, Jira, Confluence

More: https://flaviusmarc.com/experience/continental.html

### Research Intern, AI and Computer Vision, Erasmus MC (research internship)

Jun 2023 to Sep 2023, Rotterdam, Netherlands

- Trained a U-Net in PyTorch that segments the internal carotid artery in X-ray angiography (test Dice 0.875), and built a Python application that turns DICOM series into four perfusion maps without manual steps. The work contributed to the perfDSA paper; research report in the Utrecht University repository (PDF: https://studenttheses.uu.nl/bitstreams/3574e4be-62cf-4f79-bf8d-7c6f484d67d6/download).

More: https://flaviusmarc.com/experience/erasmus-mc.html

### Research Intern, literature review, UMC Utrecht (research internship)

May 2023 to Jun 2023, Utrecht, Netherlands

- Reviewed 51 sources on monitoring congestive heart failure with seismocardiography (accelerometer-recorded chest vibrations): its link to filling pressures, wearable sensors and machine learning. Published in the Utrecht University repository (PDF: https://studenttheses.uu.nl/bitstreams/0fd2ba79-f719-4c33-8266-ee005a40d213/download).

More: https://flaviusmarc.com/experience/umc-utrecht-literature-review.html

### Research Intern, UMC Utrecht (research internship)

May 2022 to Feb 2023, Utrecht, Netherlands

- Diffusion MRI with an ultra-strong gradient head insert at 7 tesla, in MATLAB; co-author of the resulting paper in Magnetic Resonance in Medicine. Research report in the Utrecht University repository (PDF: https://studenttheses.uu.nl/bitstreams/3c37665f-c9b5-4c2f-a468-a7003718d25d/download).

More: https://flaviusmarc.com/experience/umc-utrecht.html

## Projects

### AI micro-course learning platform

Mar 2025 to Mar 2026. Own project.

Generates a personalised short course and an AI tutor for each learner. I built it alone, end to end: LangGraph orchestration, LangSmith tracing, FastAPI services, a PostgreSQL database, the front-end and deployment on Google Cloud.

Built with: LangGraph, LangChain, LangSmith, FastAPI, PostgreSQL, GCP

More: https://flaviusmarc.com/projects/ai-learning-platform.html

### Alex, a learning assistant for Slack

around Jan to Feb 2026. Own project.

Alex turns everyday work questions in Slack into short, personalised learning paths. I built it alone: a Slack bot on a LangGraph state machine, with a FastAPI backend, Supabase and deployment on Google Cloud Run.

Built with: LangGraph, Slack Bolt, FastAPI, Supabase, GCP

More: https://flaviusmarc.com/projects/slack-learning-assistant.html

### Medical note classifier

Mar 2025. Own project.

Reads medical notes from PDFs and sorts each one into one of five clinical specialties. I fine-tuned Bio_ClinicalBERT, built a containerised REST API around it, and made a version that runs entirely in your browser.

Built with: Python, PyTorch, Transformers, ONNX, FastAPI, Docker

More: https://flaviusmarc.com/projects/medical-note-classifier.html

### Agentic hub: from OpenClaw to Hermes

Feb 2026 to now. Own setup, running around the clock on my own server.

My own always-on AI agents, on a private server and reached through Telegram. It started on OpenClaw in February 2026 and moved to Hermes in June 2026. An appointment agent books and cancels real appointments with an email inbox of its own, and a fitness agent coaches me after every gym session I log.

Built with: Hermes Agent, OpenClaw, OpenRouter, Telegram, AgentMail, Browser automation, Gmail

More: https://flaviusmarc.com/projects/agentic-hub.html

### Question answering over PDF pages with ColPali

Feb 2025. Own project.

Ask questions about PDFs, charts and tables included, without OCR: every page is indexed as an image, the pages that answer the question are found by late interaction, and Gemini answers from them. I built the web application around the open-source ColPali engine.

Built with: Python, ColPali, RAG, FastAPI, PostgreSQL, Svelte

More: https://flaviusmarc.com/projects/colpali-retrieval.html

### Carotid artery segmentation in X-ray angiography

Jun 2023 to Sep 2023. Research internship at Erasmus MC, Rotterdam.

A deep learning model that segments the internal carotid artery in X-ray angiography, and a Python application that uses it to make perfusion images of the brain without manual work. Built during my research internship at Erasmus MC; the work contributed to a paper published in 2025.

Built with: PyTorch, U-Net, Medical imaging, Python, DICOM

More: https://flaviusmarc.com/projects/carotid-artery-segmentation.html

### Image processing library in C++

Nov to Dec 2021. MSc coursework, PMI programming course.

A library for 3D medical images written from scratch in C++: file readers, an image class and a family of filters, including a median filter built on the algorithm from Sonka's textbook. You can run it on a brain scan in your browser.

Built with: C++, Image processing, WebAssembly demo

More: https://flaviusmarc.com/projects/cpp-image-processing-library.html

### Calculator in C++

Nov to Dec 2021. MSc coursework, PMI programming course.

An expression calculator in C++, built up from the exercises in Stroustrup's Programming: Principles and Practice. It cuts what you type into tokens and evaluates them with a small grammar. You can use it in your browser and watch the tokens appear.

Built with: C++, Tokens and grammars, WebAssembly demo

More: https://flaviusmarc.com/projects/cpp-calculator.html

### Camera calibration and 3D voxel reconstruction

Early 2022. MSc coursework, Computer Vision course at Utrecht University, in a group of two.

Three connected assignments in C++ with OpenCV: calibrate the cameras, rebuild a person in 3D from four video views, then tell four people apart by the colour of their clothes.

Built with: C++, OpenCV, K-means, Gaussian mixture models

More: https://flaviusmarc.com/projects/3d-voxel-reconstruction.html

### Clothing classifier on Fashion-MNIST

Spring 2022. MSc coursework, Computer Vision course at Utrecht University, in a group of two.

Small convolutional networks trained from scratch to recognise ten kinds of clothing, compared one design change at a time. The best one runs in your browser on real test pictures.

Built with: Python, TensorFlow, Keras, CNN

More: https://flaviusmarc.com/projects/fashion-mnist-cnn.html

### Brain tissue segmentation in MRI

Dec 2021 to Jan 2022. MSc coursework, image processing course, in a team of three.

Splits a brain MRI slice into background, skull, cerebrospinal fluid, grey matter and white matter, using only thresholds and morphological operations in MATLAB.

Built with: MATLAB, Thresholding, Mathematical morphology

More: https://flaviusmarc.com/projects/brain-mri-tissue-segmentation.html

### Prostate segmentation in 3D MRI

Early 2023. MSc coursework, Capita Selecta in Medical Image Analysis at TU Eindhoven, in a team of five.

Outlines the prostate in a new 3D MRI scan automatically by reusing outlines an expert drew on 15 earlier patients: atlas-based segmentation with rigid, affine and B-spline registration in Elastix. Team project; I built the registration pipeline, which reached a mean Dice of 0.83 on unseen patients against 0.75 for the team's U-Net.

Built with: Python, Elastix, SimpleITK, NumPy, SciPy, Image registration, PyTorch (team)

More: https://flaviusmarc.com/projects/prostate-mri-segmentation.html

### Action recognition in images and video

Spring 2022. MSc coursework, Computer Vision course at Utrecht University, in a group of two.

Recognises what people are doing, first in still photos and then in video clips, with a residual network, transfer learning and optical flow. TensorFlow.

Built with: Python, TensorFlow, Keras, Transfer learning, Optical flow

More: https://flaviusmarc.com/projects/action-recognition.html

### Predicting chest shape after scoliosis surgery

Sep 2021 to Apr 2022. MSc Team Challenge in Medical Imaging 2021 to 2022, in a team of five.

Predicts what a scoliosis patient's rib cage will look like after spinal fusion surgery, from the CT scan taken before it: the patient is matched to earlier patients on six deformity indices, then the scan is reshaped onto the best match's post-operative scan with B-spline registration in Elastix. Team project; I built the registration and the main pipeline.

Built with: Python, Elastix, Image registration, MeVisLab, 3D Slicer

More: https://flaviusmarc.com/projects/scoliosis-surgery-outcome-prediction.html

## Skills

### Programming languages

- Python (used at or in: Xyall, Bitwise Europe, Erasmus MC, CART-Tech, AI micro-course learning platform, Alex, a learning assistant for Slack, Action recognition in images and video, MSc Medical Imaging)
- C and C++ (used at or in: Continental, Bitwise Europe, CART-Tech, Image processing library in C++, Calculator in C++, Camera calibration and 3D voxel reconstruction, BSc Applied Electronics, MSc Medical Imaging)
- C# (used at or in: Xyall, Bitwise Europe)
- Java and Kotlin (used at or in: Bitwise Europe)
- MATLAB (used at or in: CART-Tech, UMC Utrecht, Brain tissue segmentation in MRI, BSc Applied Electronics, MSc Medical Imaging)

### AI and agentic systems

- LangGraph (used at or in: AI micro-course learning platform, Alex, a learning assistant for Slack)
- LangSmith (used at or in: AI micro-course learning platform, Alex, a learning assistant for Slack)
- Agentic workflows with LLMs (used at or in: AI micro-course learning platform, Alex, a learning assistant for Slack)
- Retrieval-augmented generation (RAG) (used at or in: Question answering over PDF pages with ColPali)
- Cursor (used at or in: Xyall)
- Also: Claude Code, RAG, Multi-agent systems, Embeddings, Tool calling, LangChain, crewAI, AutoGen, Make.com, MCP clients, Context engineering

### Machine learning and computer vision

- PyTorch (used at or in: Erasmus MC, Carotid artery segmentation in X-ray angiography, Medical note classifier)
- TensorFlow and Keras (used at or in: Clothing classifier on Fashion-MNIST, Action recognition in images and video)
- OpenCV (used at or in: Camera calibration and 3D voxel reconstruction)
- Classical machine learning: K-means, Gaussian mixture models, nearest-neighbour matching (used at or in: Camera calibration and 3D voxel reconstruction, Predicting chest shape after scoliosis surgery)
- Image segmentation (used at or in: Erasmus MC, Carotid artery segmentation in X-ray angiography, Brain tissue segmentation in MRI, Prostate segmentation in 3D MRI)
- Image registration (used at or in: Prostate segmentation in 3D MRI, Predicting chest shape after scoliosis surgery)
- Medical image processing (used at or in: CART-Tech, Erasmus MC, UMC Utrecht)
- MRI and diffusion imaging (used at or in: UMC Utrecht)
- SciPy and pandas (used at or in: CART-Tech, Erasmus MC)
- 3D mesh generation (used at or in: CART-Tech)

### Web, backend and cloud

- REST APIs (used at or in: Xyall, Bitwise Europe)
- Flask (used at or in: Xyall)
- FastAPI (used at or in: AI micro-course learning platform, Alex, a learning assistant for Slack, Medical note classifier, Question answering over PDF pages with ColPali)
- PostgreSQL (used at or in: AI micro-course learning platform, Alex, a learning assistant for Slack, Question answering over PDF pages with ColPali)
- Google Cloud Platform (used at or in: AI micro-course learning platform, Alex, a learning assistant for Slack)
- Front-end (used at or in: AI micro-course learning platform, Question answering over PDF pages with ColPali)
- Linux and virtual machines (used at or in: Bitwise Europe)
- Also: Docker, Git, CI/CD, Jira, MySQL, WebSockets, RabbitMQ, Supabase

### Embedded, desktop and hardware

- Bare-metal firmware on ARM Cortex-A53 (used at or in: Continental)
- Lauterbach TRACE32 debugging (used at or in: Continental)
- CANoe and CAN (used at or in: Continental)
- UML design and unit testing (used at or in: Continental)
- Code and design reviews (used at or in: Continental)
- Qt (used at or in: CART-Tech)
- DSP boards and microcontrollers (used at or in: BSc Applied Electronics)
- LabVIEW and VHDL (used at or in: BSc Applied Electronics)
- Also: Arduino, Electronics hardware design, Automotive radar and camera systems

### Spoken languages

- Romanian: mother tongue
- English: professional
- Dutch: learning, A2 level
- Portuguese: conversational
- Spanish: conversational

## Education

### MSc in Medical Imaging

Sep 2021 to Dec 2023, Utrecht University and TU Eindhoven, Netherlands

Image processing, computer vision and deep learning for medical images. The degree ended in three pieces of research of my own: a major research project at UMC Utrecht, a minor research project at Erasmus MC and a literature review. Each report is published in the Utrecht University thesis repository, and both research projects later contributed to journal papers.

### BSc in Applied Electronics

Oct 2015 to Jul 2019, Politehnica University Timisoara, Romania

Faculty of Electronics, Telecommunications and Information Technology. Microcontrollers, DSP boards and electronics hardware design; final exam 9.63 out of 10. My diploma project was a device that remotely monitors a patient's finger temperature, heart rate and blood oxygen level. The final year overlapped with my job at Continental.

### Courses and certificates

- AI Agentic Design Patterns with AutoGen, DeepLearning.AI, Aug 2024
- Multi AI Agent Systems with crewAI, DeepLearning.AI, Jul 2024
- AI Agents in LangGraph, DeepLearning.AI, Jul 2024
- English for Academic Purposes, Babel Language Courses, Jan 2022

## Publications

### Journal papers

- perfDSA: Automatic Perfusion Imaging in Cerebral Digital Subtraction Angiography. R. Su, P. M. van der Sluijs, F.-G. Marc, F. te Nijenhuis, and 7 others. International Journal of Computer Assisted Radiology and Surgery, 2025. https://link.springer.com/article/10.1007/s11548-025-03359-4
- Feasibility of strong diffusion encoding and fast readout using a plug-and-play head gradient insert at 7 T. G. C. Arends, E. Versteeg, A. De Luca, F. Marc, and 5 others. Magnetic Resonance in Medicine, 2025. https://onlinelibrary.wiley.com/doi/10.1002/mrm.30613

### Research reports (Utrecht University thesis repository, open access)

- Improved diffusion MRI with an ultra-strong gradient head insert. Major research project at UMC Utrecht, Utrecht University, 2023. Tests whether a gradient coil that plugs into an existing 7 tesla scanner can make diffusion MRI sensitive to myelin water. Covers a cable layout that cuts the magnetic force on the supply cables, and a preprocessing pipeline that corrects the image distortions. PDF: https://flaviusmarc.com/reports/major-research-project-diffusion-mri.pdf Record: https://studenttheses.uu.nl/handle/20.500.12932/43702
- Automatic Cerebral Perfusion Imaging using Deep Learning for Digital Subtraction Angiography. Minor research project at Erasmus MC, Utrecht University, 2023. A deep learning model that segments the internal carotid artery in X-ray angiography, and an application built on it that produces brain perfusion images without manual annotation. PDF: https://flaviusmarc.com/reports/minor-research-project-perfusion-dsa.pdf Record: https://studenttheses.uu.nl/handle/20.500.12932/45405
- Feasibility of monitoring congestive heart failure with seismocardiography: a literature review. Writing assignment, Utrecht University, 2023. A literature review of seismocardiography, which records the chest vibrations of the heart with accelerometers, as a way to monitor patients with congestive heart failure. PDF: https://flaviusmarc.com/reports/writing-assignment-seismocardiography.pdf Record: https://studenttheses.uu.nl/handle/20.500.12932/44205
