Nice talk. Concise and providing the just the right amount of information. massive thank you for using animations in your slides it helped tremendously with your flow. Trying the github repo as we speak.
I've done and presented a project like this with more transparency over 5 years ago, and completed it within a few weeks time. The only concern there was, was with polysemy (word with multiple parts-of-speech). It really helped to condense the information down and easily see implications across the documents.
we have a few episodes covering this topic. For example the going meta episodes ( th-cam.com/play/PL9Hl4pk2FsvX-5QPvwChB-ni_mFF97rCE.html ) or th-cam.com/users/liveNbyxWAC2TLc
I think what you do there is pre index the vector database, and before sending the request, you preload the top n chunks. And most likely optimizing the knowledge graph by limiting the amount of tokens per chunk to the most optimal number for different tasks.
@@MrGara1994 if that the case this might be a dumb question 😅, but if we are using vector to get top n chunks then is there any different with doing kg or normal vector search?
Thank you very much for this helpful and inspiring presentation!
Nice talk. Concise and providing the just the right amount of information. massive thank you for using animations in your slides it helped tremendously with your flow. Trying the github repo as we speak.
Thank you
I've done and presented a project like this with more transparency over 5 years ago, and completed it within a few weeks time. The only concern there was, was with polysemy (word with multiple parts-of-speech).
It really helped to condense the information down and easily see implications across the documents.
chunking is one of the most steps to build a stable RAG flow, KG will change the RAG Game
Surprised you didn't use the Matrix movies instead :D
Can you give me more details on how to use LLM to convert text to knowledge graph?
we have a few episodes covering this topic.
For example the going meta episodes ( th-cam.com/play/PL9Hl4pk2FsvX-5QPvwChB-ni_mFF97rCE.html ) or th-cam.com/users/liveNbyxWAC2TLc
Quick question let say we are working with maybe 100s of files to create graph, would'nt it be too costly to use llm?
That's the real question
@@MrRubix94 Any idea on how we can solve it?
No idea. I have yet to dive into the subject myself.@@Manu-m8w6m
I think what you do there is pre index the vector database, and before sending the request, you preload the top n chunks. And most likely optimizing the knowledge graph by limiting the amount of tokens per chunk to the most optimal number for different tasks.
@@MrGara1994 if that the case this might be a dumb question 😅, but if we are using vector to get top n chunks then is there any different with doing kg or normal vector search?
can you provide information regarding seed from URI for azure storage seed provider
Does anyone develop application for production in this way? What about ontology?