Building Nu - Challenges in International Expansion - Nubank DS & ML Meetup #71

We are bringing together Nubankers from some of our offices around the World (Brazil, Mexico and Colombia) to talk about the incredible work we are doing as Nubank goes international. They will tell us about how we can exchange ideas and knowledge to scale DS between countries, and also bring some very interesting challenges on how to monitor models!

Agenda:

• Scaling model monitoring - Caique Lima, Staff Machine Learning Engineer at Nubank Brazil
• Monitoring models with models: adversarial monitoring - Alejandro Reyes, Staff Data Scientist at Nubank Mexico
• Cross-pollination of DS ideas across geos - Hamadys Benavides, Sr Data Scientist at Nubank Colombia; Hector Lira, Sr Data Scientist at Nubank Mexico and Gabriel Ferreira, Data Science Manager at Nubank Mexico

This event is going to be in English. Live translation to Portuguese and Spanish will be available. The event is going to start at 7pm (Sao Paulo Time Zone) or 5pm (Bogotá and Mexico City Time Zone)

Below you can find detailed descriptions for each talk
-----

Scaling model monitoring
We will talk about the process of building a model monitoring platform, and how years of experience dealing with model monitoring data helped us to build a scalable and extensible solution

Speaker: Caique Lima joined nubank in 2018, works mainly with productivity tool for our internal customers helping people to training, deploy and monitor machine learning models at scale.
-----

Monitoring models with models: adversarial monitoring
In Machine Learning model monitoring and automated systems, one of the main challenges is drift in input data and the impact this represents in the model. There exist several methods to identify data drift, most of them based on distribution analysis; however, these are mainly univariate metrics and are focused to detect what feature is drifting but not to explain why it is drifting. In this talk, we are going to show an approach that monitors and identifies feature drift in a ML model using another model. We will discuss some properties of this approach like the intuition behind it, the flexibility in summarizing results and the possibility it offers to explain why the input data could be drifting.

Speaker: Alejandro joined Nu Mexico in 2021 coming from a background of mobility technologies. He has worked as a Data Scientist in Nu developing Machine Learning models focused to improve our credit risk decisions.
-----

Cross-pollination of DS ideas across geos
How to best leverage our experience to spread our data science know-how across geos? Two poles are at the core of this question. Should we replicate solutions across geos? Or should we build from scratch using local knowledge? What is the right balance? During this talk we'll discuss three different cases with different answers to these questions and their outcomes.

Speakers:
Hamadys Benavides is a Senior Data Scientist at Nu Colombia since 2021. She works with machine learning and the intersection between machine learning and causal inference in credit decision models. She developed one of the first version of the Colombian credit acquisition model and currently, she is leading the developing of the spending elasticity model.

Hector Lira joined Nubank almost 2 and a half years ago being one of the two first data scientists hired in Mexico. Started his journey in the Fraud team developing a fraud ID machine learning model and then moved to the credit team, developing the first customer credit risk and spending elasticity models.

Gabriel Ferreira joined Nu 3 years ago, coming from a background of embedded systems and image processing. Has worked as a Machine Learning Engineer, since then, mostly in Platform and MLOps-related projects. Recently moved to the Customer Excellence Mexico team and started serving as Associate Data Science Manager.