Faiçal AMMI SAID

Particularities of the role Designed a multi-agent chatbot on Azure to support customer service and streamline internal requests, in close collaboration with Product and Data teams (Agile context). Responsibilities • Developed conversational flows with LangGraph (LangChain Multi Agent framework). • Deployed solution with FastAPI Rest API, Docker containers, and Azure DevOps pipelines. • Tested with real conversations and LLM-as-a-Judge (Ragas) evaluations. • Monitored performance via MLflow KPIs with automated tracking. Results • Increased response time for support team reducing tickets that are automatically handled by chatbot. • Increased productivity and better tracking of recurring requests. • Time savings through automated performance monitoring.

Particularities of the role Development and production deployment of passenger traffic forecasting models in an urban transport setting. Responsibilities • Preprocessed and structured traffic data with reproducible data pipelines. • Built and benchmarked ML time series forecasting models, with experiment tracking and model versioning. • Deployed models to production via GitHub Actions CI/CD pipelines, adding monitoring to detect performance drift. Results • Delivered robust, production-grade forecasting models. • Enabled better scheduling and capacity planning through actionable insights. • Directly contributed to optimizing transport schedules and resource allocation.

Particularities of the role Development and automation of ETL pipelines to support analytics and reporting initiatives. Responsibilities • Participer à l’ingestion de données depuis SQL Server vers Azure. • Aider à charger et organiser les données dans Azure Data lake. • Contribuer à nettoyer et préparer les données pour la BI. Results • Streamlined workflows, reducing manual effort. • Delivered actionable insights via dashboards and reports. • Improved query performance, speeding up analytics and decision-making.

Particularities of the role Development of a semantic segmentation pipeline for agricultural disease detection, targeting pixel-level accuracy and applied research innovation. Responsibilities • Built a pipeline reaching 95% IoU for leaf lesion segmentation. • Created a new dataset ensuring class balance and quality. • Designed and ran 7+ PyTorch training pipelines, testing multiple architectures. • Drove model innovation with Vision Transformers, optimizing scalability and performance. Results • Delivered highly accurate segmentation, providing a solid foundation for research. • Influenced methodological choices and subsequent projects. • Enhanced internal expertise in deep learning for agriculture.