Building Generative AI application with MongoDB Atlas, Amazon Bedrock, Langchain, and Streamlit

Welcome to our repository, showcasing a reference architecture for building a production-ready Generative AI application. Designed with startups in mind, this architecture leverages:

• MongoDB Atlas: A scalable, cloud-based database for efficient storage and retrieval of unstructured data, crucial for training and running AI models.
• Amazon Bedrock: A managed service for building, training, and deploying machine learning models at scale, streamlining generative AI workflows.
• Langchain: A tool for seamless integration of complex natural language processing pipelines, enhancing the application's ability to understand and generate human-like text.
• Streamlit: A framework for rapid development of interactive, real-time data-driven web applications.

Key Features

• Production-Ready Chatbots: Set up sophisticated AI-driven chatbots using customized Docker images.
• Scalability and Flexibility: Leverage the cloud for scalable and flexible deployments.
• Comprehensive Guide: Step-by-step instructions for developers to build and deploy their AI solutions.

Prerequisites

• MongoDB Atlas Cluster and Database - Deploy a free MongoDB Atlas Cluster
  • Optional: For access to free Atlas credits, apply to the MongoDB startup program
• AWS Account AWS Free Tier

Reference Architecture

MongoDB Atlas Setup

Load MongoDB Atlas Sample Dataset

• In MongoDB Atlas Console, from your Cluster menu select Load Sample Dataset

Vector Search Index Creation

• Navigate to Data Services > Your Cluster > Browse Collections > Atlas Search

• Select movies collection from sample_mflix database and copy and paste the JSON snippet below.

  {
  "fields": [
    {
      "numDimensions": 1536,
      "path": "eg_vector",
      "similarity": "cosine",
      "type": "vector"
    }
  ]
}

• Select Next

• Confirm Index Creation by clicking on Create Search Index

• Observe index being created

• Once the index is in status Active, it is ready for use.

AWS Setup

Creating a Secret in Secrets Manager

• You can skip this section and create the secret via CLI in the next section
• In AWS Console navigate to Secret Manager and click on Store a New secret

• Choose Other type of secret then Plaintext tab and provide the connection string of the MongoDB Atlas Cluster. We have created the connection string in the previous lab.

• Proceed to the next page where you supply secret name as workshop/atlas_secret

• Continue to other pages without any changes to store the secret

• Confirm the secret is created by observing it listed on the page. Note: you might need hit refresh button to reload the secrets.

Enable Amazon Bedrock Model Access

• In this step we need to enable access to several models
• In AWS Console, navigate to Amazon Bedrock|Model Access.

• Click on Manage Model Access and select all models under Amazon. Click on Request model to proceed.

• Observe Access status to be Access Granted

Setup EC2 Instance

• Start an EC2 Instance

• Connect to the EC2 Instance

• Install Git and Docker

sudo yum install git
sudo yum install docker
sudo service docker start
sudo usermod -aG docker $USER

Relogin once to get the above setting effective.

• Configure AWS CLI

aws configure

Generative AI application setup

• Clone the repo

git clone https://github.com/mongodb-partners/AWS_MongoDB_Generative_AI.git
cd AWS_MongoDB_Generative_AI/

• Install Python Dependencies

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Create Embeddings

• Run the following code in terminal and wait for the program to finish:

python ./bedrock_atlas_vector_search_streamlit/create_embeddings.py

• Now observe the vector containing embeddings created in MongoDB Atlas. Note: because we are not updating the full dataset you might need to filter the records by supplying this filter expression: {"eg_vector":{"$exists": true}}

Run Search to verify

• In terminal run the following command to perform vector search

 python ./bedrock_atlas_vector_search_streamlit/query_atlas.py

• Verify that program returns search results:

• Next, we run the program that adds a generative feature.

 python ./bedrock_atlas_vector_search_streamlit/llm_atlas.py

• Based on the retrieved description, we are now generating a description for a new movie.

Run Streamlit app

• Before we can run the streamlit app, we need install ECS Copilot CLI by running the following commands

 sudo curl -Lo /usr/local/bin/copilot https://github.com/aws/copilot-cli/releases/latest/download/copilot-linux    && sudo chmod +x /usr/local/bin/copilot    && copilot --help

• Install the Application and Environment using ECS copilot

copilot init

• Select the Application name : atlasgenai

• Select the Load Balanced Web Services

• Name the service must be frontendas the manifest files is configured for frontend.

• Enter the custom path for Dockerfile

• Give yes to "Would you like to deploy an environment?"

• Name of the environment must be dev, as the manifest files is configured for dev.

• Ensure the environment is setup successfully

• Then the copilot execute the docker file, pushes the image to the repository and creates the infrastructure. Ensure all the mentioned are created successfully.

• Update the role name created by the copilot with the pattern "atlasgenai-dev-frontend-TaskRole' set the permission for the role to read security and bedrock

aws iam attach-role-policy --role-name atlasgenai-dev-frontend-TaskRole-<XXXXXX> --policy-arn arn:aws:iam::aws:policy/AmazonBedrockFullAccess

aws iam attach-role-policy --role-name atlasgenai-dev-frontend-TaskRole-<XXXXXX> --policy-arn arn:aws:iam::aws:policy/SecretsManagerReadWrite

• Open the external URL form the copilot output - refer to the above screenshot for reference - in a separate browser tab. You should see the app loaded

• Type several keywords to generate a new movie description. for example romantic comedy sunshine. You can add or remove keywords to generate new descriptions.

Clean up

copilot app delete

Services & Cost Estimations(Approx)

• MongoDB Altas -- Free Cluster
• AWS Account Free tier
• AWS EC2 t2.micro - free
• AWS S3
• AWS Secret Manager
• AWS ECS
• AWS IAM Role
• Network balancer
• Cloudwatch

Summary

• Congratulations! You have deployed an app that performs semantic search in MongoDB Atlas and mashes up several descriptions into one!