Quickest RAG Setup — AWS Bedrock Knowledge Base in 5 minutes
Create your LLM chatbot API with RAG ready for your product without fine-tuning, yes in 5 minutes.
Yes just upload a bunch of company document files(excel, pdf, docx) to S3 bucket and follow this setup, you will have company LLM(think ChatGPT, Claude) chatbot API in an instant.
Before you go on, If you haven’t read about AWS Bedrock yet, please check it out here
Ready? Let’s get into it.
Caution: In this exercise, DO NOT FORGET to delete Opensearch serverless(in the section “Delete Opensearch Serverless”) after you are done with the knowledge base and the setup, Follow this guide and you are good to go.
If you forget to delete Opensearch serverless, check the “I forgot to delete Opensearch serverless” section
Limitation
- You need to send an email to AWS support if your region is not eligible to use Knowledge Base.
- At the time of this writing, I have to select the
us-east-1region to try out AWS Bedrock entirely. Because it is in the “preview” phase
Let’s get into it
- Go to AWS Bedrock Console
By searching “Bedrock” in a search box
2. Open the menu on the left side scroll to the bottom and select “Model access”
3. Click “Manage model access”
4. Select “Claude”, you may need to send a use case for this but it takes a few seconds to do.
If you want you can request access to other models too, wait a few minutes you can try it on the playground. But that’s not why we are here, let’s move on to the Knowledge Base
5. Open the left-hand side menu and click on “Knowledge Base”
6. Click Create Knowledge Base
7. Input the Knowledge base name
8. Select “Create and use a new service role”
9. Input “Data source name” and select S3 Buckets that you put all “knowledge” files in
10. Configure vector store
11. Select “Quick create a new vector store — Recommended”, this is going to cost some money, so follow the last step to make sure you don’t lose too much money
12. On the left hand side is the steps diagram, and now we are ready to review and create the Knowledge base
13. Click “Create knowledge base”
14. It should take a few minutes to create a Vector database in Amazon Opensearch
15. There should be green alert when it’s done.
16. Let’s focus on the box “Test knowledge base” on the right hand side
17. Now you can click “Sync data source” and then try to ask the question with the input box below
18. And we are done! If you are done playing around with the knowledge base, let’s continue the cleanup and make sure we delete Opensearch Serverless.
Delete Opensearch Serverless
- Go to the search bar, search for “Opensearch” and Click on “Amazon Opensearch Service”
2. Open the left-hand side menu and look for and click at “Serverless”-> “Dashboard”
3. Once you are at the dashboard, look under “Collections” and delete the one that was used for Bedrock Knowledge Base. Click “Delete” and wait until the collections is delete.
I forgot to delete Opensearch serverless
Well, Well, Well, you come back here after a few days/weeks, because your bill is skyrocketing, because, you guessed it, you forgot to delete Opensearch serverless. The solution is simple. Open a support ticket in the AWS console and be honest about your situation. I got a cost reduction 100% after chatting with the support for a couple of hours as well.
Next Step
The next step would be creating a Knowledge base with other data sources such as Pinecone and Aurora which will cost a lot less and try out AWS CLI and AWS SDK to interact with the Knowledge base, follow for more!
