🎤 Talks
01:12 Retrieving with Haystack by David Batista
Slides: https://drive.google.com/file/d/1yl8ChujJFigvPr-f8Mbf3b3Yg5TI6ryi/view?usp=sharing
One of the main components in a RAG system is the retriever, which fetches relevant documents from a document store. In this talk, we will revisit different retrieving techniques, ranging from the classical ones that originated within the Information Retrieval community and moving on to recent techniques that leverage LLMs to efficiently index and retrieve relevant documents. We will see the main differences between these techniques and how some can be combined to achieve better results. We will also refer to the Haystack library to show how to use them in practice.
33:21 NLP-as: an NLP analytics system that helps you develop insights into the limitations of your NLP engine by Oren Matar
Slides: https://drive.google.com/file/d/1F1lEkNJMN9beVnm3AcMVwP8Cwidi_N6G/view?usp=sharing
As NLP engines integrate into numerous applications, assessing their robustness and limitations remains a challenge. This session introduces an innovative, model-agnostic method that leverages LLMs to generate robust test datasets, complete with labels for linguistic issues. This approach enables precise root cause analysis of failures, providing deep insights into NLP engine strengths and weaknesses - essential for improving performance, especially with black-box models like LLMs and generative AI.