DIGITAL TWIN
TwinFlux — EV Energy Intelligence
Physics-calibrated EV simulation (SUMO + HVAC) achieving ±0.6% Wh/km across 5 vehicle profiles. AutoML + PINN SoC prediction API with real-time MLflow tracking.
AI Engineer & Data Scientist from Marrakech, Morocco, with a foundation in mathematics and computer science. Published researcher with work in Sensors (MDPI) and Springer LNNS. I love building end-to-end systems, and there's nothing quite like the satisfaction of watching them work exactly as intended.
I build ML systems that hold up outside the notebook, physics-informed neural networks grounded in real equations, LLM pipelines that automate complex document workflows, and digital twins that process telemetry in real time. My work lives at the overlap of research rigour and production reliability: prototype fast, then harden it until it scales.
I believe the next generation of AI will be built at the intersection of physical laws and data-driven learning, not choosing between the two. My goal is to contribute to that frontier: models that are not just accurate, but interpretable, physically constrained, and trustworthy enough to deploy in high-stakes environments.
When I'm not building systems, I'm out running, logging miles and collecting race medals when the opportunity shows up. I hike whenever terrain allows and travel as often as I can, drawn to new places, landscapes, and the perspective that comes with being somewhere unfamiliar.
PyTorch
TensorFlow
Keras
Hugging Face
YOLO
OpenCV
OCR
Claude
Gemini
MLflow
Apache Airflow
FastAPI
Docker
Grafana
Prometheus
PostgreSQL
MongoDB
InfluxDB
MySQL
MinIO
Redis
CARLA
SUMO
React
Next.js
JavaScript
HTML5
CSS
Tailwind CSS
Python
SQL
GCP
AWS
Oracle Cloud
Huawei CloudPhysics-calibrated EV simulation (SUMO + HVAC) achieving ±0.6% Wh/km across 5 vehicle profiles. AutoML + PINN SoC prediction API with real-time MLflow tracking.
CNN & RNN (GRU, Bi-LSTM) models on the DEAP dataset reaching 91–94% accuracy in valence/arousal classification. Published in Sensors, MDPI.
Physics-Informed Neural Network for EV battery State-of-Charge estimation, embedding the Coulomb Counting ODE directly into the loss function — <2% prediction error with guaranteed physical consistency.
Conditional GAN trained on 5,000+ annotated frames to generate synthetic agent trajectories for RL environment augmentation — 93% target-detection accuracy and sim-to-real transfer.
Full-stack AI system classifying 7 Moroccan genres (Gnawa, Chaabi, Andalusian, Raï, Imazighn, Rap, Pop) from audio clips. LSTM on MFCC features, 95% test accuracy on 5,000+ clips, containerised Flask API with React frontend.
Research on augmented-reality applications fostering student engagement and active learning in STEM environments. Published in Springer LNNS.
Whether it's a job opportunity, a research idea, or just a question — drop me a message and I'll get back to you.