Data Scientist

About the role

Join the Performance team to advance our advertising outcomes through sophisticated modeling. This role focuses on developing prediction models and bidding strategies for streaming TV, an area with significant growth potential and unique data challenges. You will contribute to a rapidly evolving field, shaping the future of cross-device attribution and identity graphs.

Key facts

What you'll do

• Enhance our unified multi-task model for predicting user visits, purchases, and other key outcomes.
• Create shared data representations to improve learning across various conversion funnels.
• Manage the full lifecycle of model development, from initial concept to deployment in production.
• Troubleshoot and resolve issues in production models, including gradient problems, convergence, and data drift.
• Collaborate with Performance engineers on serving and bidding infrastructure.
• Develop new features from raw data signals, including household-level features from our identity graph.
• Prepare the modeling system to integrate new and richer data sources.
• Optimize models to achieve measurable advertiser uplift, focusing on real-world business impact.
• Deploy changes to production, evaluate their effects, and make adjustments based on actual results.

Requirements

• Practical deep learning experience deployed in a professional setting.
• Strong proficiency in Python, with a preference for PyTorch (TensorFlow or JAX are also acceptable).
• Experience identifying and resolving model failures such as data leakage, bias, calibration, and distribution shifts.
• Capability to develop algorithms from the ground up and understand their internal workings.
• Demonstrated history of connecting modeling work to specific, improved business key performance indicators.

Nice to have

• Experience with very large datasets, including billions of impressions, events, or user records.
• Deep learning application to tabular data and sparse user representations.
• Background in click-through rate (CTR) or conversion rate (CVR) prediction, recommendation systems, or identity graphs and cross-device attribution.
• Familiarity with production machine learning tools like ONNX export, orchestration (Dagster), and inference serving (Triton).

Skills & tools

• Python
• PyTorch (TensorFlow, JAX)
• Deep Learning
• ONNX
• Dagster
• Triton

Practical notes

This is a hybrid role, requiring team members to be in our Paris office three times a week. Benefits include full health insurance via Alan, meal vouchers via Swile, an annual company offsite, and quarterly in-person tech syncs for engineering and product teams. The interview process includes a recruiter screen, manager interview, technical interview, calibration interview, and reference checks.