I build software. Mostly the kind that learns from data: computer vision, NLP, classical ML. Based in the Netherlands, working remotely through my own company.
Production ML system for a plant robotics lab. Drop in a petri dish scan of five seedlings, get back root tip coordinates with confidence scores. The hard part wasn't the model. Condensation and water droplets show up as more pixels than some of the smallest roots, so most of the work went into filtering that noise reliably. Residual Attention U-Net, Dijkstra pathfinding, patch-based inference. REST API, PostgreSQL, Next.js dashboard, Airflow weekly retraining, Azure ML. F1 0.88.
Real-time Dutch Sign Language off a webcam. EfficientNet-B0 primary, landmark MLP fallback below 70% confidence. 15-frame smoother, Dutch dictionary suggestions, speedrun mode. Standalone exe.
Paste a YouTube URL, get per-sentence emotion labels. Whisper or AssemblyAI transcription with speaker diarization, MarianMT translation, fine-tuned transformer for 7-class classification.
Real-time causal graph of stock relationships rendered in a custom 3D JavaScript environment. Tigramite infers directed causal links rather than plain correlations. Instead of watching price charts, you see which stocks are actually driving which others and how that structure shifts in real time.
CNN classifier for damaged and diseased coral from underwater images. Built for eventual drone footage integration, giving conservation teams automated monitoring that scales without manual review.
Benchmarked multiple traditional ML methods to predict patient stay duration from admission features.
Research project exploring whether music listening habits can serve as a proxy indicator for mental health outcomes, giving policy makers a data-backed signal without needing direct clinical data.
Data Science & AI student at Breda University of Applied Sciences. I've built a production CV system for a plant robotics lab, a real-time Dutch sign language app, and an NLP pipeline that goes from YouTube URL to per-sentence emotion labels.
The hard problems are usually not the model. It's the messy data underneath it. I work across computer vision, NLP, and classical ML. Outside of studying I take on freelance work through my own company, Tito Technologies.
Fully remote, project-based. If there's a problem involving data and a model, reach out.
Remote, project-based, no long-term commitment. Tell me what you're building.
Send a message →