My work, projects
and experience

I'm 19, currently studying B.S. Engineering Science at KU Leuven. I develop software and projects in my free time.

02

Experience

Two-week internship at Approved Automotive in Dubai, a premium luxury car dealership with 50+ employees. I rotated across CRM, sales, procurement, marketing, reception and media, seeing how each part of the business connected. The CRM stood out as the company's operating layer, linking customers, cars, tasks and teams. Seeing that system in motion strengthened my interest in building businesses, not just software.

Projects

Smart doorbell system

Raspberry Pi Zero 2 W, WebRTC, React Native, Go

Built a privacy-conscious smart doorbell with live video, two-way audio, motion recording, and mobile app control. The interesting part was making a real product-like system work under tight hardware, network, and security constraints. University project.

  • Designed a React Native app for live viewing, playback, and doorbell events.
  • Built WebRTC live streaming with two-way audio between the device and app.
  • Wrote a custom Go daemon to coordinate recording, playback, camera/audio sources, and stream forwarding.
  • Used a NoIR wide camera with IR lighting so the system could see in the dark.
  • Designed the Supabase backend with PostgreSQL tables, storage buckets, and RLS policies.
  • Chose the Raspberry Pi Zero 2 W because its compute and memory limits made the project a more realistic engineering challenge.

Escape room symbol scanner

Computer vision, Raspberry Pi, YOLOv8s

For a high-school escape room assignment, I built an AI-powered puzzle where players drew symbols on paper, placed them under a camera, and unlocked a code only if the right sequence was recognized. It turned the assignment into a working computer-vision system running locally.

  • Fine-tuned a pretrained YOLOv8s classification model on a custom four-class image dataset.
  • Expanded a small dataset with synthetic augmentation such as rotations and distortions.
  • Ran inference locally on a Raspberry Pi instead of relying on a cloud API.
  • Evaluated the model with validation accuracy, a confusion matrix, and unseen test images.
  • Reached 100% validation accuracy on the small validation set.
  • First tried building a PyTorch classifier myself, both from scratch and with pretraining, which taught me why dataset size and transfer learning mattered here.