Extending Knowledge of LLM

We will talk about the necessary base for creating RAG applications using large language models.

At the beginning, we'll have a brief look at the available LLM market and ways to expand their knowledge. Then we'll talk about specific user scenarios, among them — ones using YandexGPT. Afterwards, there is going to be a discussion of the strengths and potential limitations of these models. Then, we will review available ways to expand knowledge in the cloud — through the Datasphere and through the building of RAG (Retrieval-Augmented Generation).

The second part of the talk is going to be about analyzing the high-level architectures of RAG applications. We will cover the main components of such systems, the principles of their interaction and the key stages of information processing. An important part of the talk will be the technical details of the implementation of such applications: an overview of available databases (vector databases, managed PostgreSQL, OpenSearch), embedding methods, and tokenization. In conclusion, we will discuss methods for evaluating the quality of the RAG application.

This talk will be continued by our colleague Dmitry Soshnikov, who during his workshop will show in practice how to build a RAG application using LLM.