Projects

Predicting Renewable Energy Potential in Europe with PySpark on AWS EMR

Designed and executed a large-scale distributed data engineering pipeline using PySpark on Amazon EMR to process the NASA POWER climate dataset (Europe subset). Transformed raw climate data into partitioned Parquet format, trained a cloud-phase classification model with Spark MLlib to predict solar energy potential, performed SQL-based analysis via Amazon Athena, and monitored cluster performance with Amazon CloudWatch.

Armada: ETA Prediction Model, Replacing Google Maps API

Built and deployed a production-ready ML system that replaced reliance on Google Maps API with an in-house ETA prediction service, directly reducing operational costs. Compared multiple algorithms (Linear Regression, Random Forest, XGBoost), deployed the best model as a Flask API on AWS EC2 with Gunicorn + Nginx, and automated weekly retraining via AWS Glue to keep the model accurate with fresh data.

Chfamma?: Real-Time Tunisian Data Platform

Built a live data platform displaying real-time sports, weather, and tourism data on an interactive Tunisian map. Engineered a full scraping pipeline (BeautifulSoup, Requests) feeding into MongoDB, refreshed hourly via AWS Glue jobs, exposed through a Flask REST API on AWS EC2, and visualized with React + Leaflet.js. Deployed frontend on Vercel with automated backend refresh.

Armada: Driver & Company Performance Analytics

Developed a complete analytics system during the Armada internship that transformed raw delivery operations data into actionable business insights. Built interactive dashboards for driver KPI tracking (acceptance rate, response time, ETA compliance) and company-level performance metrics, empowering managers with data-driven tools for workforce optimization and operational decision-making.

Weather Forecasting with LSTM

Designed and trained a Long Short-Term Memory (LSTM) neural network to predict daily weather trends based on historical meteorological datasets. Covered the full ML lifecycle: data cleaning, feature engineering, model training, validation, and visualization of predictions vs actual outcomes.