I'm a Master's student in AI & Data Science (double degree between FSDM Fès and Université Sorbonne Paris Nord), passionate about building scalable data systems and transforming raw data into valuable insights.
I enjoy working across the whole data lifecycle: ingesting data in real time, modeling it properly, orchestrating reliable pipelines, and turning it into forecasts and ML-driven decisions.
🎯 Goal: Become a skilled Data Engineer by building real-world projects and developing reliable, scalable data architectures.
| 🔄 Data Pipelines | ⚡ Streaming | 🏛️ Warehousing | 🔁 CDC | ☁️ Big Data |
|---|---|---|---|---|
| ETL / ELT end to end | Real-time & distributed | Dimensional modeling | Debezium + Kafka | Cloud & lakehouse |
The Total Commits card shows commits for the current year, and the streak card shows total contributions with the current and longest streak.
Debezium Kafka Spark Delta Lake SeaweedFS (S3) DuckDB dbt Great Expectations Airflow Docker
A change-data-capture pipeline that streams every insert, update and delete from PostgreSQL through Debezium and Kafka (KRaft) into a Delta Lake on S3 with Spark Structured Streaming (Bronze → Silver). Data is then loaded into DuckDB, modeled with dbt, validated with Great Expectations, and orchestrated with Airflow. Handles live schema evolution and exactly-once recovery after a kill -9.
flowchart LR
A[(PostgreSQL<br/>wal_level=logical)] --> B[Debezium Connect]
B --> C[Apache Kafka]
C --> D[Spark Structured Streaming]
D --> E[(Bronze Delta)]
E --> F[(Silver Delta)]
F --> G[(DuckDB Warehouse)]
G --> H[dbt Models]
H --> I[Great Expectations]
J{{Airflow}} -.orchestrates.-> D
J -.-> H
J -.-> I
Spark Kafka Cassandra HDFS Python Scala
An end-to-end ingestion pipeline built on Kafka and Spark Structured Streaming, feeding Cassandra for low-latency market queries on the MASI index. Historical tick data is archived in HDFS and used for parallelised ARIMA forecasting.
flowchart LR
A[Market Data Source] --> B[Apache Kafka]
B --> C[Spark Structured Streaming]
C --> D[(Cassandra)]
C --> E[(HDFS Archive)]
E --> F[ARIMA Forecasting]
D --> G[Low-latency Queries]
Python Scikit-learn StatsBomb
An ML pipeline that estimates shot quality (xG) from StatsBomb event data using location, body part, and play-pattern features. It matched the official StatsBomb benchmark across three classifiers.
Python PyTorch Federated Learning
A privacy-preserving federated/split learning pipeline that predicts depression indicators from head images. CNN models are trained across distributed clients with federated averaging, so no patient data is ever centralised.
Python LightGBM LSTM cvxpy yfinance
A LightGBM and LSTM ensemble for financial time-series forecasting, with walk-forward backtesting and mean-variance optimisation via cvxpy to build risk-adjusted portfolios.
| 🎓 Université Sorbonne Paris Nord | Double-Degree Master's in Data Science (In progress) |
| 🎓 FSDM Fès | Master's in Data Engineering & Data Science (Sep 2025 – In progress) |
| 🎓 FSDM Fès | Licence in Mathematics & Computer Science (2022 – 2025) |
| 💼 AGH Data Agency Holding SA | Data Analyst Intern (Jul 2026 – Aug 2026) |
🇲🇦 Arabic (Native) • 🇫🇷 French (Fluent, C1) • 🇬🇧 English (Fluent, C1)

