Passionate about technology and learning. I ship AI and data systems end to end, from an LLM assistant running in a live factory to TinyML on microcontrollers, backed by years of high-tech electronics and embedded engineering. The demos below run live on your device.
A retrieval-style assistant grounded in my experience. Ask how I build intelligent systems, what I've deployed, or where I come from.
AI Engineer · Intelligent systems & automation · Eindhoven, NL
Passionate about technology and learning.
Two tracks, one mindset: intelligent systems that run in the real world. Click any card for the details.
From models to monitored production systems.
Deployed an LLM (DeepSeek-R1) as a chatbot in a live electronics factory, with n8n workflow automation for ticketing and process orchestration around it.
Why it matters: operators get answers at the line without waiting on engineers. The hard parts were production-grade integration: prompt and context design, guardrails, and wiring it into the existing ticketing flow so it automates work instead of adding a toy.
Built a Python pipeline that extracts and processes measurement data from production servers, surfaces patterns across large test datasets, and feeds prediction models, cutting failure-analysis time significantly.
Why it matters: test data was sitting unused on servers. The pipeline turns it into process-capability stats and trend insight that drive data-backed decisions on test strategy.
End-to-end pipeline from raw IMU sensor data to an optimized CNN running on-device: preprocessing, feature extraction, training, and low-latency edge inference on Raspberry Pi.
Why it matters: the whole loop runs on the edge, no cloud round-trip. Quantization and model surgery to hit real-time latency on constrained hardware is exactly where the electronics background pays off.
CNN trained from scratch (98% test accuracy) served with TF.js, a RAG-style CV assistant, and a weekly ETL pipeline on a CI cron, all deployed automatically from git.
Why it matters: nothing here is a screenshot. The models run client-side in WebGL, the data refreshes itself weekly, and every push redeploys through CI. The portfolio is itself a shipped system.
From schematic to firmware to a working system.
Designed the hardware and firmware from the ground up: BMS and power electronics, sensor fusion and real-time control on FreeRTOS/ESP32-S3, ML-assisted PID tuning, and onboard OpenCV vision.
Why it matters: one system spanning analog power, real-time firmware, and applied ML, designed modular so autonomous features drop in without redesign.
Develop automated test tooling for complex boards from leading high-tech OEMs: NI LabVIEW/TestStand test development, PCBA design in Altium, and C++/Python firmware.
Why it matters: these testers gate real products shipping to semicon, scientific and industrial customers. Reliability and clear failure diagnostics are the product.
Full-cycle analog design for precision signal detection in low-SNR environments: filtering, phase-sensitive detection, and gain staging.
Why it matters: recovering microvolt signals buried in noise is unforgiving; every dB of noise floor is earned in topology, grounding, and component choice.
HDMI video pipeline on a PYNQ-Z2 FPGA, a SIMO buck-boost converter with integrated LDOs, ESP32 face detection, and a solar-tracking USB-C PD station.
Why it matters: built for the learning, kept for the range: gateware, switching power, and embedded vision are different disciplines, and the curiosity to cross them is the point.
Seven years from physics in Agadir to AI systems in Eindhoven, traced as one line. Scroll to draw it; filter by what you care about.
Automated test development for high-tech PCBAs, plus the data layer on top: analytics pipelines, failure prediction, a production LLM assistant, and n8n automation.
Custom drone platform from the ground up: BMS and power electronics, FreeRTOS on ESP32-S3, sensor fusion, ML-assisted PID tuning, onboard OpenCV vision.
Hands-on help and translation for newly arrived refugees, plus the Wake Up Your Mind project for young people.
Graduated with honours, specialisation Electronic Systems. Thesis: TinyML movement recognition from IMU sensor data, deployed on-device.
Real-time IMU movement recognition on Raspberry Pi: custom PCB, embedded C acquisition, full ML pipeline, optimized TinyML model for low-latency inference.
Automated testing workflows and data-driven performance evaluation for high-power LED drivers (400W to 1800W).
Treasurer of a TU/e photography association; vice president of a village association in Morocco working on safe water, solar power, and education.
Reverse engineered a VESC-based e-bike motor controller: firmware structure, control algorithms, hardware architecture.
Low-noise, high-sensitivity lock-in amplifier for precision detection in low-SNR environments: full-cycle analog design.
Pre-master track alongside the bachelor, deepening the mathematical and systems foundations.
PR for the honours community: open days, events, and getting students excited about going beyond the curriculum.
Where it started: physics fundamentals, circuit theory, and the curiosity that led from Morocco to the Brainport region.
Client-side inference, WebGL-accelerated, nothing uploaded. The MNIST CNN was trained by me (98% test accuracy); gesture and detection use SOTA pre-trained models.
Live OWID data, refreshed weekly by a CI pipeline: trend, heatmap, and an interactive map across 28 European countries.
The kit I reach for, and how deep it goes. Honest levels, no tens.
Python 
PyTorch 
TensorFlow 
Keras 
scikit-learn 
OpenCV 
Hugging Face 
MLflow 
Databricks 
PySpark 
pandas 
NumPy 
PostgreSQL 
D3.js 
TypeScript 
FastAPI 
n8n 
Docker 
Git 
GitHub Actions 
C / C++ 
ESP32 / STM32 
Arduino 
Raspberry Pi Verifiable where possible: credential IDs included, links open the issuer's verification page.
Based in the Brainport region, Eindhoven. Open to AI, ML, and data roles where intelligent systems meet the real world.