Providing our AI Agents with external knowledge through RAG is crucial for a majority of AI applications. But generally, a basic RAG implementation is only going to be 50-75% accurate when retrieving the correct information from your knowledgebase. If you want to build real AI solutions, that just won’t cut it.
I've been researching all the best strategies for making RAG more accurate (and scalable for millions of documents), and enhancing RAG with knowledge graphs has really stuck out to me, specifically using an incredible open source library called LightRAG. LightRAG not only vectorizes your documents like traditional RAG, it also builds up a knowledge graph that connects concepts, entities, and ideas together to take the contextual understanding of your AI agent to a new dimension.
In this video, we'll build a LightRAG implementation that you can use as a starting point to build some VERY powerful RAG AI agents.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Check out Graphiti (free & open source) if you want to build RAG applications that work with real-time data:
https://github.com/getzep/graphiti
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
LightRAG GitHub repo:
https://github.com/HKUDS/LightRAG
Link to my LightRAG implementation (with the ChromaDB version too):
https://github.com/coleam00/ottomator-agents/tree/main/light-rag-agent
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
00:00 - Introducing LightRAG - Accurate & Scalable RAG
02:02 - LightRAG Overview & GitHub Repo
05:07 - LightRAG vs. "Traditional" RAG & GraphRAG
05:53 - Get Started with LightRAG in 2 Minutes
07:44 - Our LightRAG AI Agent
09:27 - Graphiti - Real-Time Knowledge Graphs
10:54 - Testing Our LightRAG vs ChromaDB AI Agents
13:29 - Building Our LightRAG Knowledge Graph
16:48 - LightRAG AI Agent Code Deep Dive
20:09 - Covering the ChromaDB "Traditional" RAG Agent
21:36 - Final Thoughts
22:34 - Outro
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Join me as I push the limits of what is possible with AI. I'll be uploading videos at least two times a week - Sundays and Wednesdays at 7:00 PM CDT!