Arnaud Hureaux

Comprehensive modelization of the dynamic relationship between Bank Relationship Managers’ commercial performance, customer advocacy measured through Net Promoter Score (NPS), and the concrete actions they perform across different stages of the client journey — including prospecting, onboarding, portfolio management, and after-sales follow-up — with the objective of identifying the behavioral patterns and managerial levers that most effectively drive both revenue growth and long-term client satisfaction.

​

DECATHLON Global Sport Retail Group (2 years)​

Context: the Customs had a process to verify 20 000 "container's PDFs" to avoid fines fromEuropean Customs Office, which mobilised 20 Full-Time-Equivalent​

As a LLM Expert, I automated the process with a workflow of 3 steps:​

Categorise => Classic NLP with TFIDF/LogisticRegression deploy on AWS Lambda​

Extract => GenAI / LLM with AWS Bedrock Anthropic models and Bedrock Batch/Airflow​

Compare => Databricks SQL/Spark Data Pipeline and AWS Lambda Python workflow​

Display => Front in Streamlit/React and deployment on EKS​

Huge success of the POC (1 month), success of the POV, proof of value (4 months) andsuccessful deployment on 12 countries (Bangladesh, India, Vietnam...), 1M of value added​

2 years of developpement, 1 alone and 1 as a Tech Lead with an Data Scientist to manage​

2 parralel projects of RAG development : to help Customs to find the best HS code for theirproduct (Pinecone & Amazon OpenSearch) + to chat with transport documentation)

Full Stack Tech Lead on DBT / GCP / BigQuery / CloudComposer / Python / C#​

Machine Learning API of price's prediction for reel estate​

Data Pipeline creation on french reel estate opendata (DVF, PCI, RNC, BDNB...)​

API creation and integration on their React website (https://dataimmosolutions.com/)