Using vector embeddings for Retrieval-Augmented Generation (RAG) is frequently demonstrated in many talks. However, embeddings have a broader range of applications including the ability to group similar items together. Log files are an ideal candidate for this approach. By clustering log entries, we can uncover common issues as well as rare conditions within complex systems.

This talk will begin with traditional techniques using term vectors and then hava a look how modern vector embedding models can enhance log file analysis. We will explore how to build such a processing pipeline, which is often an interesting and productive way to start understanding the problems of the system you're dealing with.