Mór Kapronczay, lead ML engineer at Superlinked, delivered an insightful online talk at the Data Science Festival on 23rd July, 2024, focusing on building effective RAG-powered chatbots for HR applications.

Key points of Mor's presentation included:

1. The importance of Retrieval-Augmented Generation (RAG) in creating business value with Large Language Models (LLMs)
2. A shift in focus from text generation quality to improving retrieval performance
3. Techniques for combining embeddings from both numeric and text data into a single vector
4. A practical demonstration of building a high-performing RAG system for HR policy inquiries

Mór emphasized that while public attention often centers on the quality of generated text, developers can achieve more significant improvements in RAG performance by optimizing the retrieval process. This approach is particularly valuable for datasets containing mixed data types, common in many business applications.

Check out the VectorHub article for detailed instructions and try it for yourself in the GitHub repo.

Watch the full talk from the Data Science Festival below.