So clear and simple compared to other libraries for building genetic pipelines. Intuitive and feels like it should've been in Hugging Face libraries from the start. Makes other libraries seem overly complex and unnecessary. Easy to create an LLM engine with just a callable class. You can build any structure, with complexity only from yourself, not the library. Not surprising from Hugging Face, just like how fine-tuning models with HF library is intuitive and easy. Love a simple, powerful library that doesn't over-abstract. This is the way. Thanks for sharing.
Thanks for the video! it would be interesting to see a multiagent and routing approach with 2 sources like a vector store for rag and a sql db, each one with their agents
It would have been interesting to see if you would have used GPT-4o as the LLM engine in the traditional RAG method to compare it with the agentic RAG response.
Unfortunately, using agents in the loop with take longer than standard RAG since it has to make additional calls to the LLM and do retrieval again. Over time you can cache queries and responses for faster retrieval.
thanks for the video. I would like to analyze PDF studies of several hundred pages and make summaries to extract insights. The problem is that I can't copy/paste the pdf into GPT because it goes beyond the context window. Can I use RAG to do this use case? The RAG seems to be designed more for answering specific questions from a knowledge base than for synthesizing documents.
Agent has the ability to do multiple passes of retrieval if it's not able to find the info in the first pass. If you add this to the system prompt, I will just run once and can't repeat the process with reasoning and Planning.
It really depends on your use case. GraphRDF is currently ten to twenty times more expensive. Also, depending on the type of data and the type of query, it could be useful for you or not. It also increases lag by a very substantial margin. I have not found any startups or ideas implementing graph-lag effectively and usable yet. If you do, please keep me in the loop.
Checkout the Advanced RAG course here: prompt-s-site.thinkific.com/courses/rag
thanks.. Can you become the ISHOWSPEED of AI. also are you based in USA or Subcontinent?
@@criticalnodecapital haha, that would be a good achievement :D I am based in the USA.
So clear and simple compared to other libraries for building genetic pipelines. Intuitive and feels like it should've been in Hugging Face libraries from the start. Makes other libraries seem overly complex and unnecessary. Easy to create an LLM engine with just a callable class. You can build any structure, with complexity only from yourself, not the library. Not surprising from Hugging Face, just like how fine-tuning models with HF library is intuitive and easy. Love a simple, powerful library that doesn't over-abstract. This is the way. Thanks for sharing.
Yeah, really like their implementation. Clean and straightforward.
Cool idea! When a course comes out focused on csv or databases, then I'll be there!
Thanks for the video! it would be interesting to see a multiagent and routing approach with 2 sources like a vector store for rag and a sql db, each one with their agents
It would have been interesting to see if you would have used GPT-4o as the LLM engine in the traditional RAG method to compare it with the agentic RAG response.
In your video the agentic rag takes about 4 times longer (15 sec). Is there a way to speed up agentic rag?
Unfortunately, using agents in the loop with take longer than standard RAG since it has to make additional calls to the LLM and do retrieval again. Over time you can cache queries and responses for faster retrieval.
Agentic RAG + Knowledge Graph would be bad ass. Someone steal my idea, please. 😂 🙏
working on it!!!!
@@severian42 any updates?
Any updates?
@@severian42 Any updates?
thanks for the video.
I would like to analyze PDF studies of several hundred pages and make summaries to extract insights.
The problem is that I can't copy/paste the pdf into GPT because it goes beyond the context window.
Can I use RAG to do this use case?
The RAG seems to be designed more for answering specific questions from a knowledge base than for synthesizing documents.
YES OFCOURSE!
Why do I need an agent when I can add the agent description as a system prompt
Agent has the ability to do multiple passes of retrieval if it's not able to find the info in the first pass. If you add this to the system prompt, I will just run once and can't repeat the process with reasoning and Planning.
Great video can we use GROQ API for the LLM?
It depends upon the prompt message you give.. Yes we can utilize GROQ models for simpler agentic RAG process
OP, Under what circumstances might you have duplicate chunks? Is it becuase two files that are same with differnt names?
Yes, that happens a lot. In big datasets, there can be duplicates.
I dont remember the last time I had and issue with hallucinations.
how does this compare to graph rag?
It really depends on your use case.
GraphRDF is currently ten to twenty times more expensive. Also, depending on the type of data and the type of query, it could be useful for you or not.
It also increases lag by a very substantial margin. I have not found any startups or ideas implementing graph-lag effectively and usable yet.
If you do, please keep me in the loop.
THANKS!
too technical. where are friendly user interfaces websites/apps?