Raphael Gallula

• Processed and consolidated Mercedes’ multi-channel marketing and financial data to build and analyze the complete customer conversion funnel, developed locally (VS Code) and deployed on Azure Blob (Pandas, NumPy, Azure, GCS). • Designed and developed a Mercedes-Benz France Customer 360 DataMart (~20 years of historical data) by consolidating multiple data sources (customers, vehicles, fleet, orders, and deliveries). Built data pipelines for integration, standardization, matching, deduplication, and monthly updates to ensure a reliable and up-to-date customer repository, used for scoring models and ML-based targeting. (Azure Blob, Python, SQL Server) • Modeled Aldi’s email engagement data, creating KPIs and cohort analyses to track user retention and behavioral trends. • Cleaned, structured, and organized Armonia’s client data, developing AI algorithms to ensure optimal data quality. Applied machine learning to identify opportunities and promote additional services to existing clients. (Python, streamlit, API Sirene )

Monthly automation of financial reporting across 20+ countries, including the processing of historical files: design and implementation of a Control-M–orchestrated workflow for automated file collection, standardization of Excel reports, and data mapping using Python scripts to unify formats into a single structure, followed by data cleaning and processing in Dataiku, and the creation of interactive dashboards in Tableau. ---- Master’s thesis: Development of an embedded voice assistant for vehicles, combining local speech recognition, on-device LLM, contextual dialogue management, dynamic web access, and appointment scheduling to assist drivers.

Subject : "Leveraging AI for dependability and safety data analysis" I worked on integrating CATIA with large language models (LLMs) to automate the generation of failure modes (FMs) for functions within subsystems. I optimized result visualization, improved precision, and contributed to comprehensive FMEA (Failure Mode and Effects Analysis) of entire, complex systems with multiple subsystems. By developing algorithms to assess the plausibility of each failure mode, I enriched the LLMs with data from standards, public accident reports, and private data. This allowed me to propose reliability and safety studies based on in-depth experience, while laying the groundwork for future full integration of the tool within CATIA. Python, CATIA, Pixtral AI

Intern Position: Data Scientist Subject: "How to Query a Database Using Natural Language" Designed and developed an LLM agent using the OpenAI, LangChain, Streamlit, and Snowflake frameworks. Created a user-friendly graphical interface in Python that enables users to interact with a database using natural language, whether in written or spoken form. The interface also includes data visualization capabilities. Ensured complete protection and confidentiality of data, with no data leaks.

Military service.