Activity
Mon
Wed
Fri
Sun
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
Dec
What is this?
Less
More

Memberships

JorrDaar AD-venture

1.7k members • Free

AI Automation Society

208.5k members • Free

3 contributions to AI Automation Society
RAG Assistant for my customers based on my own websites
A few days ago, I came across WAHA (Whatsapp API for N8N) (we use Whatsapp for customer chats) and by then had already seen videos about using Telegram as an assistant, and Nate's and other YTers videos on RAG. But all of them (including Nate) had examples on doing RAG using documents in Google Drive. I don't use Google Drive at work and I don't create documents (at least not yet), that would be beneficial to my customers. And so, I thought about creating a workflow that did this: A workflow that took whatsapp messages like, "what's the price?", "what is covered in your service?", "are you open now?", "how do I download X?" and provide answers automatically from the content of my websites. And for the past week, I have been trying to get it done. I am proud to say that I finally got the RAG from websites working. I am sharing that workflow here. This is my first draft and I haven't really bothered with aesthetics (naming of nodes and responses). Will do that once I know that everything is working A-OK. Things I still need to do: - Create snippet in my wordpress websites to trigger n8n webhook on page, post, products and courses creation. - Create snippet in my wordpress websites to trigger an n8n webhook for page, post, products and courses updation. - Update the workflow (fetch page nodes after webhook trigger) to take the url passed by the webhook. - Create my Whatsapp Assistant workflow to answer queries on behalf of my team. You can download and use it freely. Just make sure that if you enhance it in anyway, to share it forward and to this community. Oh, and I am no expert in n8n. I just started using this last month and from all the searches, chats on n8n official community and asking people like Nate, it would be clear that I literally have no expertise in this. I just mashed up a lot of stuff from here and there to get this working. And used a lot of AI chats. So, please don't ask me to explain to or troubleshoot for you. Tip: Claude seems to be better at figuring out n8n issues than Chatgpt, at least in some areas.
1 like • Dec '24
@Jelord Chien My pleasure...
0 likes • Dec '24
@Simon Scrapes that's a good idea... i am right now trying to figure out why my chat agent won't answer basic questions even though i clearly see the text in the database... but once I get that part working, i'll implement your suggestion. :)
Personal assistant through whatapp
I like to know is it possible to create a personal assistant over whatapp where i will instruct it to help me find restaurant and give me the review and url of the google map To remind me at 3pm to call this person and it will make a note in my calendar To draft and email to inform that person i have make the following change
0 likes • Nov '24
Hey. check out WAHA. I just started using it.
New Video: Step by Step: RAG AI Agents Got Even Better
In this step-by-step tutorial, I walk you through a more automated approach to creating RAG agents—making the process smoother than ever. I test out using Postgres for memory and Supabase for the Vector Database. While there's still room for optimization, this method takes a big leap forward from past builds on this channel. Thanks for all of the support lately everyone, road to 2k members in AI Automation Society!! Let me know your thoughts on this one. As always, the workflows are attached to this post.
0 likes • Nov '24
@Nate Herk hey. thanks for replying. here are some screenshots. IMG 1 this is what I am loading into the vector database. the content has the content, but I am passing title, url and rowID to metadata. I changed the workflow yesterday. I was anyway capturing all the content of all the pages to regular supabase table, and thought I might as well use the record/row id of the content to identify. so, the rowID i am passing to the vector store is the rowID of the entry in the regular table. it works fine till here. (at least it did until last night when I broke something upstream. but i'll fix it. no biggy). this was just FYI to show you what I am sending to vector store. IMG 2 this is a part of the workflow that checks if the page is already in the regular table. if it isn't, then it does the normal extraction and saving to regular table and vector store. this branch works fine. it's the other branch I am having trouble with. the first supabase node (Supabase, update row) is the one that just updates the regular table with any new content from the website. I figured, why delete and create new, when I can just update the record with new content. the second supabase node (Supabase3, get rows) is the one that searches the vector store for any records that have a metadata with rowID = rowID of the updated regular table row. IMG 3 [this is supabase3 node from above] this is where I get stuck. I tried a version of the string you had in your example "metadata->>file_id=like.*{{$json.file_id}} (i wrote it from memory and hence could be wrong), but that didn't work. I can't use like because it's a number and at some point those numbers might repeat. for example, 2 -- 23 -- 235 -- 2350, so i can't use like {{2}} and hence used eq. operator. but I keep getting an error about "operator does not exist: jsonb ~~ unknown" sorry, I can't show the error now because last night I might have broken an earlier node and now none of it work (which I will fix sometime today).
2 likes • Nov '24
@Nate Herk Hey. thanks for trying. I fixed all my issues and got my workflow running. Will be posting a new thread sharing my JSON for your followers to use.
1-3 of 3
@ravi-warrier-1243
Had to rejoin because that was the only way to get some answers this time (hopefully). www.impresseveryinvestor.com

Active 159d ago
Joined Nov 10, 2024
Powered by