Bruno Siarry

Contributed to the Data Marketing and Data Capabilities Squads (transversal teams of 5-10 persons) and was part of the DS Chapter.

Stack: Python, GCP (BigQuery, VertexAI, GCE), Airflow+dbt, Jenkins, Tableau, Data dog, MLflow, Terraform, Bayesian modeling, Jira, Google Suite, Copilot

• In charge of developing the Marketing Mix Model (MMM): measuring the impact of 10 distinct marketing channels, helping steer 2M€+ marketing budget in France. Steered Conversion Lift A/B tests with Meta and Google, on 200k live users. Proposed new timeseries and seasonality modeling.

• In charge of maintaining, monitoring and improving Cerberus, our Fraud detection program, blocking 10k accounts/day, detecting fake trips, bookings and credit card fraud. Leveraging a modern MLops stack.

• Implemented the DS pipeline for Sphinx, our text-moderation engine based on LLM embeddings, handling 10M messages/day, and saving 1 M€/y outsourced costs.

• Contributed to DS tech monitoring and organized ML practice exchange session with 10 other companies from the Parisian Tech environment.

Development of ML programs, analysis and visualization specific to the clients needs. Presentation and appreciation of the results, for clients in multiple sectors: Stack: Python (Flask, Streamlit, openCV, OCR), GitLab CI/CD, Kubernetes.

• Luxury: CV algorithm to get accurate (< 1 mm) size measurements from multiple photos, for digital sizing ofjewelry.

• Industry: program with simple user interface with OCR to accelerate specifications document reading, and automate classification of products.

• Energy: deployed of a text-analysis Flask API onto Kubernetes with Gitlab CI

I was in charge of sizing, proposing and developing ML programs and analysis, and then presenting them to clients in various sectors:

• Industry: designed, developed and accompanied the adoption of a program to clas sify metadata, saving 10k+ hours of work to engineering teams.

• Insurance: contributed to an analysis of climatic data, joined with client portfolio data, to predict drought costs in France.

• Responsible AI: improving our methodology to insure fair and robust AI develop ment, leading to its certification by LNE.