Arthur – Python, LLM, AI

Sephora is a global beauty retailer offering a curated mix of prestige cosmetics, skincare, and fragrance products. Known for its strong digital presence and in-store experience, Sephora combines AI-powered personalization, exclusive launches, and a popular loyalty program to drive customer engagement and repeat business.

Arthur's main responsibilities included the following:

• developed a product search engine on Vertex AI, using LLMs, RAG, LangChain, prompt engineering and chaining techniques;
• developed and deployed a text-to-SQL chatbot to provide easy access to company data, using Cloud Run, LLMs and an agentic framework, as well as the “LLM-as-a-judge” evaluation framework;
• expanded an existing marketing mix model to new markets, automating manual data flows and improving model evaluation;
• developed multiple POCs with LLMs like an image-based skincare recommendation engine and a data quality improver.

The client was a global consulting and innovation consultancy brand within the Capgemini Group, focused on digital transformation and business innovation. It combined strategy, creative design, data science, and technology expertise to help organizations envision and build emerging business models. Operating in over 60 offices worldwide, the company works closely with C‑suite clients to prototype, scale, and implement new products, services, and sustainable growth initiatives

Main scope of responsibilities of Arthur included the following:

• saved 24 man-hours of work per week for an energy company, by automating a financial process using Generative AI;
• built multiple GenAI POCs, utilizing Streamlit and LLMs for automated content creation and chatbots;
• worked as the main technical contributor to the GCP GenAI CoE, helping sell Generative AI solutions to multiple clients.

This customer was a Paris-based consulting firm specializing in data, digital, and AI transformation solutions. The company partnered with organizations across sectors—such as finance, retail, utilities, and public services—to deliver pragmatic, intelligent systems that drive decision-making and business performance. Its service offerings included data strategy, AI product development, automation with NLP and computer vision, self-service analytics, and RPA—designed to enhance commercial impact and foster sustainable innovation

Arthur managed the following:

• reduced baseline error of revenue forecasts by 10% by using a seasonality clustering method;
• increased usage rate of discount coupons by 30%, by creating an allocation algorithm based on scoring techniques;
• led a team of 3 consultants to deliver projects on time, focusing on improving team satisfaction and productivity;
• calculated price elasticity coefficients at the SKU level to help optimize promotions;
• segmented website navigation sessions using topic modeling, to improve CRM campaign targeting.

This customer was a curated online marketplace based in Paris, specializing in second-hand furniture and home decor. It connected professional antique dealers and private sellers with buyers seeking unique, vintage, Scandinavian, and design pieces. With over 100,000 carefully vetted listings from over 2,000 sellers, the platform enabled customers to shop second-hand with confidence, focusing on quality, sustainability, and the circular economy. It operated as a trusted intermediary, handling listings, transactions, and delivery logistics across multiple European countries.

Arthur's contribution included the following:

• implemented a data warehouse on PostgreSQL and created dashboards on Google Data Studio;
• saved 40h of monthly work by automating reports with Python and Google Sheets.