Great work! One thing I would recommend is slowing down a bit. There’s a lot of useful information here, but it can be a little hard to follow. Don’t compromise on quality to shorten the length. Keep up the great work-I hope you continue doing what you do! 😊
dam spent 2 days watching your videos , great examples and including errors and ways to fix problems . just used system prompt generator on my old system prompt and it did amazing improved my reliable answers. looking forward to making my works paperwork done by ai agent :) keep up the great videos
00:03 Introducing an advanced RAG system using n8n, Supabase, and Postgres. 01:39 Managing file versions and uploads within a Superbase system. 03:17 Explaining the workflow for processing Google Docs in Project Wheel. 05:06 Understanding file handling and metadata extraction in the RAG system. 07:03 Key features of Project Wheel enhance data management and integration. 09:06 Tutorial on updating version-controlled records in a project. 11:07 Kickoff meeting rescheduled; workflow tested and project timeline confirmed. 13:08 Incrementing version numbers in file processing logic. 14:54 Overview of versioning and metadata handling in the RAG system. 16:52 Overview of automating file deletion in a project. 18:27 Instructions for accessing the resource provided in the video. Crafted by Merlin AI.
great overview Nate. I think you could take the nodes starting with your switch statement and putting it into your vector database, might be able to be its own workflow. This way you could just call that workflow in both the new and changed document flow and not repeat all those nodes twice.
Awesome video thank you! Looks for the workflow and walk through in the skool classroom, can you point me in the right direction please. Keep up the great work! Jay
Hi guys, I uploaded two files, a PDF and a Word document, to Google Drive. Why does only the Word file go through, but not the PDF? It seems the PDF can only be uploaded to Supabase if I delete the Word file.?
Hello Nate! I'm so glad I found your channel! Thank you for these many educational videos! Unfortunately, the RAG structure doesn't work for me. I've done it about 5 different ways, but it only works from the document about 10%. It's like it doesn't want to take into account that the knowledge base is there and it has to use it. I tried: pinecode, qdrant with vector database. 4 different embedding models. 3 different LLM chat models. I didn't get acceptable results with any of them. It constantly searches for the answer in its own LLM knowledge base and doesn't want to take it from the input file. Compared to this, let's say a NotebokLLM is heaven and earth. It works perfectly, but this is the unusable category for me.
Thought it would be a cool feature to see how many times a file has been updated. I’m sure there are use cases where this information is valuable. It’s also good that the agent would be able to understand the difference between original and modified docs.
@@nateherk Hi @nateherk, thank you for your response! What I mean is that I'm looking to build a SaaS platform where users can upload their files (like PDFs or documents) and interact with them through a chat interface powered by AI. I saw your RAG System 2.0, and it seems like a great starting point. My question is, can this template be adapted to include multi-tenant support (e.g., different users with isolated data) and features like file upload management and AI-powered chat? I'd love to hear your thoughts!
i see a bunch of problems with this, such as whole doc re-ingest when maybe 1 part changes. why not implement a proper KG via triples? and add date & time element for temporal memory. then you'd be able to query on global context over a timeline. without it you're basically begging to get inaccurate answers eventually
Appreciate the detailed breakdown! I have a quick question: My OKX wallet holds some USDT, and I have the seed phrase. (alarm fetch churn bridge exercise tape speak race clerk couch crater letter). What's the best way to send them to Binance?
Hi Nate, how can one reach you. I would love us to build an agent together for a potential market. I want to demo it to a potential customer, can i send you an email.
Great work! One thing I would recommend is slowing down a bit. There’s a lot of useful information here, but it can be a little hard to follow. Don’t compromise on quality to shorten the length. Keep up the great work-I hope you continue doing what you do! 😊
Thank you for your feedback! I appreciate the advice!
dam spent 2 days watching your videos , great examples and including errors and ways to fix problems . just used system prompt generator on my old system prompt and it did amazing improved my reliable answers. looking forward to making my works paperwork done by ai agent :) keep up the great videos
Cool stuff, I couldn't figure out how to get the for loop to work untill i saw this. Good Stuff!!!
00:03 Introducing an advanced RAG system using n8n, Supabase, and Postgres.
01:39 Managing file versions and uploads within a Superbase system.
03:17 Explaining the workflow for processing Google Docs in Project Wheel.
05:06 Understanding file handling and metadata extraction in the RAG system.
07:03 Key features of Project Wheel enhance data management and integration.
09:06 Tutorial on updating version-controlled records in a project.
11:07 Kickoff meeting rescheduled; workflow tested and project timeline confirmed.
13:08 Incrementing version numbers in file processing logic.
14:54 Overview of versioning and metadata handling in the RAG system.
16:52 Overview of automating file deletion in a project.
18:27 Instructions for accessing the resource provided in the video.
Crafted by Merlin AI.
Thank you for providing so much value; I’m always happy when you release a new video! 👍
great overview Nate. I think you could take the nodes starting with your switch statement and putting it into your vector database, might be able to be its own workflow. This way you could just call that workflow in both the new and changed document flow and not repeat all those nodes twice.
Great suggestion. Lots of ways to improve here. If I was to combine the logic I would just have to get creative about a few aspects but it could work.
Love this stuff Nate thank you
YES!!!! I have downloaded the .JSON Great tutorial Nate!!!
Awesome video thank you! Looks for the workflow and walk through in the skool classroom, can you point me in the right direction please. Keep up the great work! Jay
Go to RUclips Resources and look for the post associated with this video. You should see the workflow JSON there to download.
Found it, thank you
Great stuff!!! Thanks for sharing
great setup!! thanks for sharing
thank you for sharing, have a great day :-)
Great work!
Great job, thanks
Great video Nate, what is the purpose of the loop node in the workflow?
Great job
Best regards !
Hi guys,
I uploaded two files, a PDF and a Word document, to Google Drive.
Why does only the Word file go through, but not the PDF? It seems the PDF can only be uploaded to Supabase if I delete the Word file.?
Thanks Nate!
Awesome video! Is the template only in the paid skool?
No! You can find this in the free community
Would adding audio files work the same. Extracting the audio as text
Yes, you can definitely do that!
why dont you just make "version" a number (i.e. 1) instead of v1?, that way you could increment the version without using a 4o-mini call?
I think it's also a bit of weird way to do it. You add extra loading time with this and it costs money
I can't seem to find the download this project, I was added to your skool community but I can't find it.
Hey Mark! Check the tab called RUclips Resources or you can use the search bar to find the post associated with this video.
Thanks !!!!
Could yo please share match_documents function? thank you!
Hello Nate! I'm so glad I found your channel! Thank you for these many educational videos!
Unfortunately, the RAG structure doesn't work for me. I've done it about 5 different ways, but it only works from the document about 10%. It's like it doesn't want to take into account that the knowledge base is there and it has to use it. I tried: pinecode, qdrant with vector database. 4 different embedding models. 3 different LLM chat models. I didn't get acceptable results with any of them. It constantly searches for the answer in its own LLM knowledge base and doesn't want to take it from the input file. Compared to this, let's say a NotebokLLM is heaven and earth. It works perfectly, but this is the unusable category for me.
thanks for the video, quest: why do you care about updating the version of the file? v1 v2 ... for what use?
Thought it would be a cool feature to see how many times a file has been updated. I’m sure there are use cases where this information is valuable. It’s also good that the agent would be able to understand the difference between original and modified docs.
Thanks
well … this is the best that’s true thanks man 😂
Does this work with multiple files uploaded all at once?
You have to explain butter!
Haha!
thanks bro for the valuable infos❤. can you inform us on how we can build bot/rag to response customer on WhatsApp. and thanks again
I can look into making a video about WhatsApp integration! Thanks for the suggestion
Can i use it as a SaaS for my app?
This is kind of a vague question, I’m sure this template could give a good starting point! Not sure what the app idea is.
@@nateherkRag chatbot webapp, the user can login and upload his files and chat with them, how can I handle this situation?
@@nateherk
Hi @nateherk, thank you for your response! What I mean is that I'm looking to build a SaaS platform where users can upload their files (like PDFs or documents) and interact with them through a chat interface powered by AI. I saw your RAG System 2.0, and it seems like a great starting point. My question is, can this template be adapted to include multi-tenant support (e.g., different users with isolated data) and features like file upload management and AI-powered chat? I'd love to hear your thoughts!
btw using a llm to add +1 instead of programmatically... that's almost criminal, at the very least wasteful
Want to give me a hand?
@@nateherk I was able to do it with an Edit Fields node. Version is the input, output is {{ parseInt($('File Updated').item.json.version) + 1}}
This is the problem right here 😅. AI is a tool, not a problem solver. Every engineer can start to breathe again.
I'm too stupid for this and have no idea whats going on. Does this make money?
@ymeshulin how?
i see a bunch of problems with this, such as whole doc re-ingest when maybe 1 part changes. why not implement a proper KG via triples? and add date & time element for temporal memory. then you'd be able to query on global context over a timeline. without it you're basically begging to get inaccurate answers eventually
What’re your suggestions for improvement?
@nateherk exactly what he wrote lol?
@@nateherk tip: lightrag (or zep's graphiti if you have the time)
@@jd2161I think his agent wrote the comment reply
Appreciate the detailed breakdown! I have a quick question: My OKX wallet holds some USDT, and I have the seed phrase. (alarm fetch churn bridge exercise tape speak race clerk couch crater letter). What's the best way to send them to Binance?
the switch feels inefficient
Do you have any recommendations to improve it?
They are using your development on another RUclips channel without giving you the corresponding credit.
ruclips.net/video/FUc6GKAK560/видео.html
"Steal this," but it's already stolen... I'm not going to say where from, lol.. ;)
Hi Nate, how can one reach you. I would love us to build an agent together for a potential market. I want to demo it to a potential customer, can i send you an email.
Yes shoot me an email and we can discuss!
nateherk@uppitai.com