Union RAG

UnionAI-native RAG applications.

Note

If you're here for the LLM RAG workshop, go here.

Setup

conda create -n union-rag python=3.11 -y
pip install -r requirements.txt

Create the Knowledge Base

First create the knowledge base offline:

union run --remote union_rag/simple_rag.py create_knowledge_base \
   --exclude_patterns '["/api/", "/_tags/"]' \
   --embedding_type "openai" \
   --limit 100

Run the Simple RAG Workflow

For quick iteration and development, you can run the workflow on Union with:

union run --remote union_rag/simple_rag.py ask --question 'what is flytekit?'

Simple RAG Workflow Deployment

Deploy the workflow to Union with the following command:

union register union_rag/simple_rag.py

Run the Agentic RAG Workflow

Run the agentic RAG workflow with a more complex question:

union run --remote union_rag/agentic_rag.py ask \
   --question 'Write a flytekit workflow that trains an sklearn model on the wine dataset.'

Run the same workflow with feedback:

union run --remote union_rag/agentic_rag.py ask_with_feedback \
   --question 'Write a flytekit workflow that trains an sklearn model on the wine dataset.'

Slack App Deployment

Deployment instructions

Install sam cli.

We'll use bolt to create a slack bot and sam cli to deploy a lambda function that will serve as one of the backend layers for our question-answering slackbot.

1. Follow the Bolt getting started guide to create a slack app.

   • Follow the instructions to create a SLACK_BOT_TOKEN and SLACK_SIGNING_SECRET.
   • Create a UNIONAI_SERVERLESS_API_KEY using unionai create app union-rag

2. Create a secrets.txt file to store these credentials. This file is ignored by git and should look something like this:

   SLACK_BOT_TOKEN=<SLACK_BOT_TOKEN>
   SLACK_SIGNING_SECRET=<SLACK_SIGNING_SECRET>
   UNIONAI_SERVERLESS_API_KEY=<UNIONAI_SERVERLESS_API_KEY>
   

3. Export the secrets to your environment:

   export $(cat secrets.txt | xargs)

4. Create the deploy.yaml file:

   cat template.yaml | envsubst > deploy.yaml
   

5. Make sure your ~/.aws/credentials file is properly configured with your aws_access_key_id and aws_secret_access_key.

6. Login to AWS ECR. First do aws configure sso, then:

   aws sso login --profile <PROFILE>

7. Run sam build --template deploy.yaml to build the app.

8. Run sam deploy --guided to deploy the app to AWS. This will ask you a series of questions on specific values you want to use for the deployment. In the end you should see an output like this:

   CloudFormation outputs from deployed stack
   -----------------------------------------------------------------------------------------------------------------
   Outputs
   -----------------------------------------------------------------------------------------------------------------
   Key                 UnionRagApi
   Description         API Gateway endpoint URL for Prod stage for union rag function
   Value               https://xyz.execute-api.us-east-2.amazonaws.com/Prod/
   
   Key                 UnionRagFunctionIamRole
   Description         Implicit IAM Role created for union rag function
   Value               arn:aws:iam::xyz:role/union-rag-UnionRagFunctionRole-xyz
   
   Key                 UnionRagFunction
   Description         union rag Lambda Function ARN
   Value               arn:aws:lambda:us-east-2:xyz:function:union-rag-UnionRagFunction-xyz
   -----------------------------------------------------------------------------------------------------------------
   
   Successfully created/updated stack - union-rag in us-east-2
   

9. Now test your slack app by installing it in your slack workspace and typing @flyte-attendant what is flytekit?. You should see an initial response from the bot, followed by the answer to your question.