Cover Image for Ray Meetup at Criteo Paris: Building Scalable AI & Data Platforms
Cover Image for Ray Meetup at Criteo Paris: Building Scalable AI & Data Platforms
Avatar for Anyscale
Presented by
Anyscale
Advance Your AI Platform with Anyscale.
152 Going

Ray Meetup at Criteo Paris: Building Scalable AI & Data Platforms

Registration
Approval Required
Your registration is subject to host approval.
Welcome! To join the event, please register below.
About Event

Join us at the Criteo office in Paris to learn how Ray simplifies building and scaling modern AI and ML workloads. We will walk through how teams use Ray to move from a single machine to distributed compute, run training and inference at scale, and power production AI platforms.

Hear from experts at Anyscale, Criteo, and H Company on distributed training architectures, scaling reinforcement learning agents, and real-world lessons from migrating production systems to Kubernetes.

Whether you are just exploring Ray or already running workloads in production, this meetup is a chance to connect with the community, hear real-world experiences, and ask questions.

Agenda

  • 18:30 – 19:00: 👋 Doors Open & Networking

    • Check-in, grab some food & drinks, and meet the community.

  • 19:00 – 20:15: 🎤 Tech Talks

    • Ray & Anyscale for Distributed Training

      • Ali Sezer, Field Engineering Lead @ Anyscale

      • An overview of how Ray and Anyscale help teams scale from a single machine to large clusters for distributed training and inference. We will walk through common patterns, cluster management, and how Anyscale simplifies operating Ray in production.

    • Data-Intensive Distributed Training for Commerce Foundation Models

      • Arij Sakhraoui @ Criteo

      • In this talk, we share how Criteo uses Ray to run data-intensive distributed training for commerce foundation models. We focus on the training architecture itself: GPU orchestration, scaling strategies, and execution across both on-prem and cloud environments. We cover concrete implementation details, design trade-offs, problems encountered, and lessons learned.

    • Scaling Online Reinforcement Learning for Computer Use Agents

      • Hamza Benchekroun, Tony W. & Avshalom Manevich @ H Company

      • This talk presents how we build a scalable online reinforcement learning pipeline for computer use agents at H, spanning both training and inference. We’ll describe how we combine open-source frameworks (Megatron, VERL, Ray, SkyPilot) with home-grown components, and share practical lessons from migrating a production system from Slurm to Kubernetes. We will also discuss challenges around weight synchronization, learner–actor coordination, and serving fresh policies at scale.

  • 20:15 – 20:30: 🍻 Q&A & Social

    • Open floor for questions followed by networking.

About Anyscale

Anyscale, the company behind Ray open source, is a fully-managed, enterprise-ready unified AI platform. With Anyscale, companies can build, deploy, and manage all their AI use cases, bringing transformational AI products to market faster.

Try it for free here.

Join the Ray Community

Location Criteo Office, 32 Rue Blanche, 75009 Paris, France

Location
Criteo
32 Rue Blanche, 75009 Paris, France
Avatar for Anyscale
Presented by
Anyscale
Advance Your AI Platform with Anyscale.
152 Going