Build RAG on a dataset full of numbers with Mór Kapronczay, Lead ML @ Superlinked

Large Language Models (#LLMs) predominantly create value for the business world through the #RAG methodology. Even though most of the public’s attention is focused on the quality of text these models generate, in this workshop, Superlinked's lead ML Engineer argues that developers can improve RAG performance much more efficiently through retrieval.

Many RAG solutions rely on datasets that contain both numeric and text data. In this video, you will learn how to combine embeddings from these different data modalities in a single vector to build a high-performing RAG system, through an example of a chatbot for HR policies:

Check out the VectorHub article for detailed instructions - https://links.superlinked.com/vh-arti...
Try it for yourself in the GitHub repo - https://links.superlinked.com/sl-repo...

#education #AITech #RAG #Superlinked

Timestamps -
00:00 - Warmup
03:43 - Real life search is hard
05:53 - LLMs struggle with encoding numbers
10:00 - Smart Retrieval makes LLama2 7B look like GPT4
15:42 - RAG on HR policies (including text and numerical data)
35:16 - Questions


🌟 For more cutting-edge development tips and insights, be sure to like and subscribe to our channel! 🌟
🔗 Connect with us:
→ Website: https://links.superlinked.com/lp-dsftalk
→ Twitter: https://links.superlinked.com/x-dsftalk
→ LinkedIn: https://links.superlinked.com/li-dsftalk