Bettahar walid

- Migration de données MongoDB → PostgreSQL : Étude et tests comparatifs des solutions Airbyte, Airflow et Talend. Démo de workflow avec n8n pour valider le processus de migration. -Prédiction du trafic aux feux rouges – Traffic Light Prediction : Développement d’un modèle RNN pour estimer les flux de véhicules à partir de données de capteurs réels. Prétraitement et analyse des séries temporelles (TAUX / DEBIT), association des capteurs et gestion des valeurs manquantes. Mise en place du pipeline train/test/validation, évaluation et visualisation des performances (MSE, RMSE, MAE). Documentation et communication synthétique des résultats pour le suivi projet. - Intégration e-commerce automatisée avec n8n : Création et orchestration de workflows pour synchroniser commandes, stocks et données clients. Gestion des erreurs et exceptions pour garantir la fiabilité et optimiser les processus métiers. - Formation Snowflake : Acquisition des bonnes pratiques et fonctionnalités avancées de la plateforme cloud Snowflake.

During my immersion internship at GICA, I had the opportunity to develop valuable professional experience in two key areas: Cisco architecture and database management. This experience not only deepened my passion for information technology but also provided me with practical insights into the professional environment.

During my internship, I worked on a machine learning project focused on predicting fuel consumption for industrial lifting machines (e.g., reach stackers and forklifts). The goal was to build a robust, scalable pipeline to forecast fuel usage based on sensor data and operating conditions. Key responsibilities: -Processed and analyzed large-scale time-series data using PySpark within Databricks. -Trained and tracked machine learning models using MLflow, ensuring reproducibility and performance monitoring. -Deployed models in a production environment using Docker and Kubernetes on Azure Cloud. -Designed and maintained monitoring dashboards with Grafana to visualize key metrics and model performance in real-time. Tech stack: Databricks · PySpark · MLflow · Azure · Kubernetes · Docker · Grafana

During my internship at Transvalor, I worked on enhancing user interaction with the company’s flagship software, FORGE and THERCAST, by developing and fine-tuning a large language model (LLM). Using LLaMA.cpp and TorchTune, I customized the model to meet client needs and integrated retrieval-augmented generation (RAG) techniques for more precise responses. Key contributions: • Data Processing: Extracted and preprocessed textual data from HTML using Beautiful Soup. • Vector Database: Implemented ChromaDB to store and manage embeddings efficiently. • Retriever Framework: Developed a retrieval mechanism with Llama Index to ensure context-aware responses. • Inference Models: Integrated models from Hugging Face to enhance inference capabilities. • Python Scripting: Automated data extraction and processing tasks with Python.