JAX for LLM Development at NVIDIA & AI in Production at Mollie

Details
📅 Date: 07-10-2025
📍 Location: Mollie @Corso Buenos Aires, 54, 20129 Milano MI
Agenda
18:30 - Doors Open
18:50 - Welcome to PyData
19:00 - Talk 1: Large-scale JAX Development with JAX-Toolbox
Speaker: Stefano Bosisio, MLOps Engineer @ NVIDIA
19:40 - Talk 3: GaaS Station: An In-House Framework for Prod-Ready AI Agents
Speaker: Stefano Polo, Data Scientist @ Mollie
20:00 - Talk 2: Advanced Data Science for Credit Risk Modeling: The Good, The Bad, and The Defaulted
Speaker: Marta Barigozzi, Data Scientist @ Mollie
20:30 - Networking
21:00 - Join us for dinner after the event too to continue chatting!

***

Talks and Speakers
Large-scale JAX Development with JAX-Toolbox
JAX is a key framework for LLM development, offering composable function transformations and a powerful bridge between low-level compilers and high-level code.
However, moving from development to large-scale production presents common challenges, such as package dependencies and deployment configurations.
To help address these challenges, we introduce JAX-Toolbox, an open-source project that provides a robust foundation for the LLM development lifecycle.
This session will cover:

• The CI/CD architecture that provides a stable foundation for JAX-based frameworks.
• How to build GPU-optimized containers for LLM frameworks such as MaxText and AXLearn, to ensure reproducible workflows.
• Practical methods for deploying frameworks' containers on Kubernetes and SLURM-based clusters.

Stefano Bosisio is an accomplished MLOps Engineer with a solid background in Biomedical Engineering, focusing on cellular biology, genetics, and molecular simulations. He earned his PhD in Computational Chemistry from the University of Edinburgh, where he developed a strong foundation in computational methods.
After completing his PhD, Stefano transitioned into Data Science, where he began his career as a Data Scientist. His interest in machine learning engineering grew, leading him to specialize in building ML platforms that drive business success. Stefano's expertise bridges the gap between complex scientific research and practical machine learning applications.

***

GaaS Station: An In-House Framework for Prod-Ready AI Agents
Do you often get asked about the newest GenAI use cases? Or maybe you've run into a puzzling Langchain error? If so, this session is for you. You'll see how at Mollie, we tackled these challenges by building our own framework GaaS (GenAI as a Service). We'll show you how developing an in-house GenAI platform speeds up development and streamlines AI adoption across teams.
By building together concrete examples, you'll learn how a centralized REST API can make AI tools easy to use for everyone—giving each business unit a secure and efficient way to build their own AI-powered solutions. Whether you're just starting out or looking for real-world inspiration, you'll walk away with practical insights to boost your next AI project.
Stefano Polo is a Data Scientist with 4 years experience in designing, building, and deploying machine learning solutions. Currently at Mollie, he develops machine learning and Generative AI solutions that drive direct business impact. Previously, he worked as a quant at leading Italian banks, creating advanced models for risk management and trading. He holds a Master's degree in Physics, where his thesis focused on reinforcement learning for pricing financial products.

***

Advanced Data Science for Credit Risk Modeling: The Good, The Bad, and The Defaulted
Ever tried building a credit risk model when your data lives in Google Sheets and your loan statuses are about as reliable as weather forecasts?
You'll learn practical data science lessons about surviving data quality issues, the critical importance of target variable definition, adding genetics to feature selection algorithms, and how engineered transactional features can transform your predictions from "probably fine" to "we actually know what we're doing."
We’ll show how classical ML approaches like logistic regression and XGBoost remain highly effective for binary classification problems, proving that sometimes the fundamentals work better than the latest AI trends. Perfect for anyone who's ever wondered how machine learning works when your data isn't clean, your labels aren't perfect, and your stakeholders want results yesterday.
Marta Barigozzi holds a PhD in Mathematics, made the leap from algebraic geometry to data science, and currently works as a Data Scientist at Mollie specializing in credit risk and probability of default models.

🔗 RSVP now and join the PyData Milano community!