KAG Graph + Multimodal RAG + LLM Agents = Powerful AI Reasoning

[@MrRavaging]

[@GaoDalie_AI]

thank you for taking to comment, I highly appreciate that and I am happy to help you guys

[@aybaravc564]

Which one is better ? lighrag or kag ? Could you share your thoughts ?

[@GaoDalie_AI]

Depends on your specific use case. If the priority is speed and efficiency in generating coherent responses, LightRAG may be the preferred choice. However, for applications that require robust reasoning capabilities and structured knowledge integration, KAG is the best choice

[@zakariaabderrahmanesadelao3048]

This channel is at the forefront of AI. I am forever grateful to you.

[@GaoDalie_AI]

Thank you for watching , happy to help guys

[@kingking-bv3vh]

The reasoning part(SPG) is write in Java and seems black box for developer. Also, the time of inference is super high.

[@GaoDalie_AI]

I think the SPG is 100 % Python where you see Java in the GitHub

[@xXWillyxWonkaXx]

Great video.

[@GaoDalie_AI]

Thank you for taking time to watch the video and i appreciate your input

[@alkaseram]

Thanks for video, it was interesting. If I already have a knowledge graph will KAG be useful. I have a knowledge graph with taxon data, the graph contains to main node types a taxon node and a rank node. Each taxon node has a relationship (has_rank ) to rank node which hold the rank_name. Each taxon has a relationship (has_parent and has_child) to taxon node to define the hierarchy of the taxon. now lets imagine we want to provide a user with an interface that enable to make natural language queries against the graph and along with LLM. for example get me all species of taxon "taxon_abc" or get me the family of "taxon_abc". which methods would be best suits this case?

[@steffenp1113]

Very cool stuff! Is there also a pure English version of the software without Chinese?

[@GaoDalie_AI]

Thank you for watching. Unfortunately, I didn’t find any similar projects in pure English. If I do come across one, I will definitely share it with you all

[@marcusbogle5389]

great tutorial, how can we use this in other projects ti give the project better reasoning ?

[@GaoDalie_AI]

please check my substack where i have showing you how to use the code and you can start from there tracking the code and adapt it for your use case

[@AymanHaneen]

The configurations structures pls , Iam a beginner

[@GaoDalie_AI]

Please consider joining my substack where you will find all my articles: substack.com/@gaodalie

[@조바이든-r6r]

Isnt it like google deep research?

[@GaoDalie_AI]

In my opinion, because OpenKAG is a series of developments, many concepts are involved before they can come up with a final product

[@artur50]

Thx for the video, great idea, pls consider a tutaorial plus test including testing queries on a large document like financial reports from market companies or whole books on a niche topic so that an LLM has (almost) no data on…

[@GaoDalie_AI]

Thanks for watching! I’ll do my best in the next video and try it out on large documents. KAG is really cool, and their architecture and algorithm design are impressive. I’ve played around with their code, and it looks super promising!

[@artur50]

@ superb!

[@amrohendawi6007]

Terrible video on many aspects: - the demo doesn't show where this KAG solution replaces RAG - the input file and query are too simple and do not need a KAG to solve

[@GaoDalie_AI]

Hey, thank you for watching! If you rewatch the video carefully, you’ll notice that at the beginning, I addressed the RAG problem and explained how KAG solves it. I also compared it with GraphRAG, so I have covered everything about KAG. If you’re still not familiar with RAG, I recommend watching my videos, as I’ve covered many topics related to it.

[@GaoDalie_AI]

Guys if you like the project please consider to hit ⭐ Paper: arxiv.org/pdf/2409.13731 Github: github.com/OpenSPG/KAG ❣join to my Patreon: www.patreon.com/GaoDalie_AI 🙏Thank you so much for watching guys! I would highly appreciate it if you Book an Appointment with me: topmate.io/gaodalie_ai Support the Content (every Dollar goes back into the video): buymeacoffee.com/gaodalie98d Subscribe Newsletter for free: substack.com/@gaodalie FOLLOW ME : join my discord if you have any questions: discord.gg/GENrSVJN Follow me on Twitter: x.com/GaoDalie_AI Follow me on Linkedin: shorturl.at/dnvEX