Example C: Introduction to RAG with StreamNative Pulsar Functions and AWS Bedrock

In this example we use multiple workflows to create and sync embeddings, complete a similarity search, and summarize the data.

To the right we present a diagram of our workflows. We will implement this in three parts:

Part 1: Create and Sync Embeddings (AWS Bedrock with Cohere and Postgres with pgvector)
Part 2: Similarity Search (AWS Bedrock with Cohere and Postgres with pgvector)
Part 3: Summarize Results (AWS Bedrock with Meta’s Llama 3.2)

We will deploy 5 Pulsar Functions, testing each Pulsar Function as we go. The following setup should be completed before starting.

1. Complete Cluster Setup, Service Account, and pulsarctl.
2. Download code example streamnativerag1.

AWS Bedrock Setup

1. Enable Cohere Embed English v3 in Amazon Bedrock, model id cohere.embed-english-v3.
2. Enable Meta Llama 3.2 1B Instruct in Amazon Bedrock, model id meta.llama3-2-1b-instruct-v1:0 (note code uses us.meta.llama3-2-1b-instruct-v1:0)
3. Create a user in AWS IAM with the AmazonBedrockFullAccess policy.
4. Create an access key for the IAM user and create a StreamNative secret storing the access key and secret access key.

AWS RDS Postgres Setup

1. Create an AWS RDS Postgres database with a public endpoint (note that serverless AWS RDS Postgres databases cannot have public endpoints).
2. Create a password for the default postgres user and create a StreamNative secret storing the password. Here is an example of the secret name and key to use for the password.