PyData @Apiiro

This talk explores the “shift left” strategy, employing open source LLMs to detect risks during the software design phase and create clear, explanatory text to enhance understanding and suggest countermeasures. During the talk we'll discuss tools to evaluate text quality and enhance LLM performance in production.

Open Source models are used everywhere. As public entities that are downloaded from various websites they are nearly endowed with a certain level of risks. In this lecture we will cover some attack vectors that are unique for these models and describe some of the plausible attacks according to OWASP mL

Cloud traffic sniffing tools monitor traffic between services, aiming to uncover vulnerabilities. Matching this traffic to the API corresponding controllers detected in the code is challenging and crucial. This presentation will introduce an approach to achieving this match using the BERT language model with meaningful preprocessing steps.