Building an AI Insurance Assistant with Amazon Bedrock Knowledge Bases: Part 1

Today I am going to walk you guys through the process of setting up Amazon Bedrock Knowledge Bases. Amazon Bedrock is a fully managed service that allows you to build generative AI applications using your own data. In this blog, we are going to set up Bedrock with Amazon S3 as the knowledge base to create a bot that answers questions around healthcare.

Right now for this blog, I have taken 3–4 publicly available documents from an Insurance company related to their healthcare policies. As a prerequisite, we need to store all the documents which we want to add to our knowledge base in an Amazon S3 bucket.

1. Lets start with creating the S3 bucket, Navigate to the Amazon S3 console and Click on “Create bucket”. Make sure that you are creating the bucket in the same region as your Amazon Bedrock models. Enter an appropriate unique bucket name.

2. Leave all other settings as default, scroll to the bottom of the page and click on “Create bucket”. You should land on a page showing your bucket details. Click on the “Upload” button.

3. Click on the “Add files” button and select all the insurance files you want to add to your knowledge base from your local machine and click on “Upload” button at the bottom of the page.

4. You should be able to see the files uploaded in your S3 bucket.

5. Now, navigate to the Amazon Bedrock console. On the left-hand navigation pane, under the “Build” section, click on “Knowledge bases”.

6. Click on “Create” -> “Knowledge base with vector store”.

7. Give your Knowledge Base a name (e.g., HealthInsuranceBot-KB) and an optional description. Under the IAM permissions section, you can choose to let Bedrock create a new service role for you. Make sure that the data source type selected is set to “Amazon S3”. Click on "Next".

8. In the Configure data source setup, Give the data source a name, and in the S3 URI field, click “Browse S3” to select the bucket where you uploaded the healthcare documents in Step 3 and choose the option “Amazon Bedrock Data Automation as parser” for Parsing Strategy. Click “Next”.

9. Scroll down to the bottom of the page, leave “Chunking strategy” set as “Default chunking”.

10. Now, we need to choose an “Embeddings model”. Select an available vector model from the dropdown (such as Amazon Titan Text Embeddings v2). For the vector database, leave the default option selected to allow AWS to automatically provision an Amazon OpenSearch Serverless vector store for you. Click on “Next”.

11. Review all the configuration details on the summary page.

12. Once reviewed, scroll down and click on “Create knowledge base”. Wait a few minutes for the OpenSearch vector store and the knowledge base to be fully provisioned.

13. Once the knowledge base status shows as “Ready”, open it and locate your S3 data source. Click on the “Go to data sources” option present at top right.

14. Click on the “Sync” button to begin ingesting and chunking your policy documents into the vector database. As you can see in the screenshot below, before clicking on Sync, there is no sync history.

15. Once sync is completed, you should be able to see the previously completed sync history.

16. Once the sync is complete, go back to your “Knowledge Base Overview” page and on the top-right side of the page, you should see a “Test knowledge base” button.

17. Select a foundation model (like Anthropic Claude 3.7 Sonnet) to generate the responses. You can type in any questions here pertaining to the articles that we have added to the knowledge base. For example, you can ask, “What am I covered for?” Amazon Bedrock will frame an answer for you from the documents uploaded and also give you a reference to those documents in case you need to view the source.

18. Another example asking the bot — “What are the major exclusions in the policy?”.

In this part, we completed creation and configuration of the knowledge base, in the next part, we will integrate the Knowledge base with an Amazon Lex bot, once we configure it with Lex there are numerous integrations which can be done like integrating with Amazon Connect, Chatbots etc.

Thank you for following along and reading all the way! See you in the next blog :)