Day 5: Vector Stores Deep Dive
Learning Objectives
- - Compare all Bedrock KB vector store options (OpenSearch, Aurora pgvector, Neptune, Kendra, DocumentDB)
- - Understand hybrid search (BM25 + k-NN) and when it matters
- - Know the 'exam trigger' phrases for each vector store selection
- - Design vector store architecture for different use cases
Tasks
Tasks
0/5 completed- Read60m
Bedrock Knowledge Bases - Vector Store Options
Chunking strategies, vector store options, data source sync. Core reference.
- Read30m
AWS Prescriptive Guidance: Choosing a Vector Database for RAG
Side-by-side comparison of all vector DB options. Essential exam reference.
- Blog20m
Hybrid Search in Bedrock Knowledge Bases
8-12% improvement over keyword-only search. Know when hybrid search is the answer.
- Blog20m
OpenSearch Managed vs Serverless for KB
Key distinction: managed cluster vs serverless collection for Knowledge Bases.
- Hands-on90m
Bedrock RAG Workshop - Knowledge Base Setup Notebooks
Work through KB setup with different vector stores. Focus on OpenSearch Serverless setup.
Exam Skills
Write your understanding, then reveal the reference answer.
Hands-On Lab
Build real muscle memory with these activities.
Set Up OpenSearch Serverless Collection for Bedrock KB
Create an OpenSearch Serverless vector collection and connect it to a Bedrock Knowledge Base.
- 1 Open the OpenSearch Service console and click 'Create collection'
- 2 Select 'Vector search' as the collection type and name it 'bedrock-kb-vectors'
- 3 Configure the data access policy to allow Bedrock service principal
- 4 Navigate to Bedrock → Knowledge Bases → Create and select the OpenSearch Serverless collection
- 5 Upload a sample PDF to the S3 data source and sync — verify vectors are stored in OpenSearch
Compare Vector Store Options in the Console
Walk through the Bedrock KB creation wizard to see all vector store options and their configuration differences.
- 1 Open Bedrock → Knowledge Bases → Create knowledge base
- 2 In the vector store selection step, note all available options: OpenSearch Serverless, Aurora pgvector, Neptune, Kendra GenAI Index, DocumentDB
- 3 Click through each option and note the required configuration fields
- 4 Document the key differences: managed vs self-managed, scaling model, cost model
Scenarios
Think through each scenario before revealing the answer.
E-Commerce Vector Store Selection
- •Which vector store supports hybrid search (BM25 + k-NN)?
- •Which has built-in analytics capabilities?
- •Can the Bedrock managed store handle this level of customization?
Practice Questions
11 questions across 3 difficulty levels.
Further Reading
Go deeper into today's topics.
Deep Dive into Vector Data Stores for Bedrock KB
Compare OpenSearch Serverless, Aurora pgvector, MongoDB Atlas, Pinecone — architecture, setup, and tradeoffs.
Vector Database Comparison — AWS Prescriptive Guidance
Side-by-side comparison of all AWS vector databases for RAG: features, scaling, cost, and use case fit.
Auto-Optimize OpenSearch Vector Database
Automated HNSW tuning: ef_construction, m, ef_search, quantization — balance recall, latency, and cost.
Using Aurora PostgreSQL as a Bedrock KB (pgvector)
Setup pgvector extension, create schema, configure Data API, IAM role for Bedrock — ACID-compliant vector store.
Multi-Tenant RAG with JWT Isolation
Per-tenant OpenSearch isolation with KMS encryption for SaaS applications.
Build Bedrock KB with Amazon Kendra GenAI Index
Reuse Kendra's high-accuracy retrieval as Bedrock KB backend — cross-application index sharing.