Tuka Bade

Context:

Recruited for a newly created position, as the sole data practitioner within the company, with no existing analytical infrastructure. Reported directly to the CFO. The management team lacked a reliable, automated view of actual revenue, which was previously manually reconstructed from SAP. The objective was to build a data pipeline and a reporting application from scratch, without an existing BI middleware.

Context:

Technical documentation (manuals, guides) is written and maintained in FrameMaker (.mif) format, then manually translated using Trados Studio. The volume of documents and target languages makes this process time-consuming and difficult to scale. The objective was to assess the feasibility of an automated translation pipeline using an LLM, while preserving the structure and formatting of the source document.

Context:

The company lacked a CRM to structure customer relationship tracking. With a focus on cost control and technical independence, the decision was made to evaluate an open-source, self-hosted solution rather than a proprietary SaaS solution before any investment decisions.

Project:Creation of a consumer sensory analysis tool for perfumed products.

IFF product teams regularly test fragrances with consumer panels (France, USA) to evaluate their reception: preference, perceived intensity, olfactory profiles, and attributes. This data was previously exploited in an unstructured manner. The objective was to build a tool allowing business teams to analyze these results autonomously and compare tested products.

- Data model construction from multi-country Excel files (French/US consumer panels), merged into a single dataset

- Definition and implementation of business indicators: TOP2BOX/TOP1BOX scores, JAR (Just About Right) analysis by intensity thresholds, aggregation of attribute territories and olfactory descriptors

- Correlation analysis (Spearman) between sensory attributes and liking score, followed by training a RandomForest Regressor to deduce the importance of explanatory variables

- Development of a Streamlit application with three comparison views: product vs. fixed reference, product vs. user-selected product, inter-country comparison

- Implementation of analysis filters (product, gender, age group, visualization type)

- User authentication setup (role-based access management)

- Presentation of analyses to product teams

Data Science Instructor at Albert School (Paris campus), Bachelor and Master levels.

Six courses taught this year: Programming in Python (incl. OS & VM) and Pandas 1 in Bachelor 1 (S1), Graph Theory and Algorithms & OOP in Bachelor 1 (S2), Applied Machine Learning for Business in Bachelor 3 (S2), and Supervised Learning in Master 1 (S1).

My pedagogy covers the entire chain, from Python programming and data manipulation with Pandas, to applied machine learning and supervised learning at the Master's level, including algorithms, object-oriented programming, and graph theory. I emphasize direct practical application: students must produce usable results, not just recite concepts.