DEV Community

Cover image for Smart Search Meets LLM: AWS-Powered Retrieval-Augmented Generation
Chandrani Mukherjee
Chandrani Mukherjee

Posted on

Smart Search Meets LLM: AWS-Powered Retrieval-Augmented Generation

#aws #python #langchain #llm

๐Ÿค–๐Ÿ” Smart Search Meets LLM: AWS-Powered Retrieval-Augmented Generation

Leverage AWS-native services to build a Retrieval-Augmented Generation (RAG) agent that can answer questions based on your internal knowledge base. Use Amazon Bedrock, OpenSearch, and LangChain to create an intelligent, secure, and scalable Q&A system.

๐Ÿ› ๏ธ Tech Stack

  • LangChain โ€“ for building RAG pipeline
  • Amazon Bedrock โ€“ for LLMs (Claude, Titan, etc.)
  • OpenSearch or Bedrock Knowledge Base โ€“ for vector storage
  • S3 โ€“ to store documents
  • Titan Embeddings โ€“ for semantic search

๐Ÿ“ฆ Install Dependencies 

pip install langchain boto3 openai
Enter fullscreen mode Exit fullscreen mode

๐Ÿ” AWS Authentication
Set up AWS CLI or environment variables:

aws configure
Enter fullscreen mode Exit fullscreen mode

Or:

export AWS_ACCESS_KEY_ID=your_key
export AWS_SECRET_ACCESS_KEY=your_secret
export AWS_REGION=us-east-1
Enter fullscreen mode Exit fullscreen mode

๐Ÿ“„ Code: RAG Agent with AWS Vector Store

import boto3
from langchain.llms.bedrock import Bedrock
from langchain.chains import RetrievalQA
from langchain.vectorstores import OpenSearchVectorSearch
from langchain.embeddings import BedrockEmbeddings

# 1. Configure LLM and Embeddings
llm = Bedrock(model_id="anthropic.claude-v2", region_name="us-east-1")
embedding = BedrockEmbeddings(model_id="amazon.titan-embed-text-v1", region_name="us-east-1")

# 2. Connect to OpenSearch (or use Bedrock Knowledge Base)
opensearch = OpenSearchVectorSearch(
    opensearch_url="https://your-domain.us-east-1.es.amazonaws.com",
    index_name="your-index-name",
    embedding=embedding
)

# 3. Create RAG pipeline
qa = RetrievalQA.from_chain_type(llm=llm, retriever=opensearch.as_retriever(), return_source_documents=True)

# 4. Ask a question
query = "How does the AWS knowledge base vector search work?"
result = qa({"query": query})

print("Answer:", result["result"])
print("Sources:", result["source_documents"])
Enter fullscreen mode Exit fullscreen mode

๐Ÿ“š Optional: Use AWS Bedrock Knowledge Base
You can simplify vector storage using Amazon Bedrock Knowledge Base:

Upload your documents to S3.

Create a Knowledge Base via AWS Console.

AWS handles chunking, embeddings, and search.

Retrieve results using Bedrock APIs or LangChain's BedrockRetriever.

๐Ÿ“ Folder Structure Example

.
โ”œโ”€โ”€ docs/
โ”‚   โ””โ”€โ”€ knowledge_base/         # Local or S3 documents
โ””โ”€โ”€ rag_aws_agent.py            # Main script
Enter fullscreen mode Exit fullscreen mode

๐Ÿ”„ Whatโ€™s Next?
๐Ÿงพ Support more file types (PDF, CSV, Markdown)

๐Ÿง  Add an agent with memory and tools

๐ŸŒ Create a web interface using Streamlit or FastAPI

๐Ÿ” Use IAM roles for secure access

๐Ÿ’พ Store logs and audit trails for responses

๐Ÿ’ก Catchy Titles You Can Use
"Chat with Your AWS Docs: Build a RAG Agent with Amazon Bedrock"

"RAG-as-a-Service: Scalable AI Q&A on AWS with LangChain"

"From S3 to Smart Search: Building a RAG Bot on AWS"

"Amazon Bedrock Meets LangChain: Create Your Own Q&A Assistant"

โœ… Summary
You now have a working RAG agent that:

Retrieves content from your AWS-hosted knowledge base

Uses Titan Embeddings and Claude for smart answers

Is scalable, secure, and easily extensible

Want this turned into a full repo, Streamlit demo, or Docker package? Just ask!

Top comments (2)

Collapse
 
lucas_henry_57e7d4ec16689 profile image
Lucas Henry •

Brilliant! This could be an awesome base for building an enterprise-grade internal knowledge assistant.

Collapse
 
aiden_benjamin_52c3f6771f profile image
Aiden Benjamin •

This is a fantastic breakdown of RAG implementation with AWS! Love the clarity on how the components interact.