PyData @ Alexa Shopping (Amazon)

We would like to thank Alexa Shopping (Amazon) for hosting us PHYSICALLY

Agenda
18:00-18:30 Gathering and snacks
18:30-18:45 Welcome words from our host
18:45-19:15 Getting the most from Transformers in a latency-constrained environment| Alon Gadot + Moran Netser, ML engineers @ Alexa Shopping (Amazon)
19:15-19:45 Balanced Cross Validation for Object Detection |Jonathan Harel, Algorithm Engineer @ Ibex AI
19:45-20:00 A short break
20:00-20:30 Bluevine's journey to eradicate check fraud| Yarden Levy & Stas Khoroshevsky, Data Scientists@ Bluevine.
20:30-21:00 Real-time processing at scale with python-based lambdas| Shoham Roditi Elimelech, Software Engineer @ Memphis.dev.

RSVP now to secure your spot!
Derech Menachem Begin 121, 28th floor
Tel Aviv-Yafo, Israel

===========================================

Getting the most from Transformers in a latency-constrained environment| Alon Gadot + Moran Netser, ML engineers @ Alexa Shopping (Amazon)
Alexa Shopping deploys and serves LLMs in a latency-constrained environment. In this talk, we will discuss strategies from the ground up, go over tools, and share battle-tested out-of-the-box methodologies for deploying transformers when you need them to serve fast. We will review the trade-offs you can make to squeeze extra performance from your model while sharing tips and tricks we use that will help you minimize both latency & cost for any use case.

## ===========================================

Balanced Cross Validation for Object Detection |Jonathan Harel, Algorithm Engineer @ Ibex AI
When it comes to object detection, performing balanced cross-validation is nontrivial: We want to split the dataset such that each fold contains a roughly equal number of instances from each class. However, each image in the dataset contains multiple instances of multiple classes. In this talk, we will discuss a solution that uses linear programming, and get hands-on with “PuLP”, a package that deals with such problems.

## ===========================================

Bluevine's journey to eradicate check fraud| Yarden Levy & Stas Khoroshevsky, Data Scientists @ Bluevine.
Detecting and preventing financial fraud is hard. Doing it on one of the oldest and safest payment tools is even harder. Join us on a journey through isolation forest, convolutional layers, and boosting to see how we eradicate check fraud at Bluevine.

## ===========================================

Real-time processing at scale with python-based lambdas| Shoham Roditi Elimelech, Software Engineer @ Memphis.dev.
When doing real-time data processing, the need for scale and agility arises fast.
In this practical session, you will learn how to build the foundations to support real-time data processing at scale, agile for changes and modifications, and how to do it in a cost-effective manner using a message broker that will trigger a fleet of AWS lambdas.

***