Bedrock Knowledge Base L3 Construct

This construct provides a high-level abstraction for creating Amazon Bedrock Knowledge Bases with vector storage and data sources.

Features

Knowledge Base Creation: Creates Bedrock knowledge bases with vector configurations
Vector Store Management: Automatic Aurora PostgreSQL vector database setup
S3 Data Sources: Support for multiple S3 data sources with ingestion configurations
Vector Ingestion: Advanced parsing and chunking strategies
Auto-Sync: Optional Lambda-based automatic synchronization with S3 changes
IAM Management: Automatic creation of required IAM policies and roles
Logging: CloudWatch log group setup with delivery configuration

Usage

import { BedrockKnowledgeBaseL3Construct } from '@aws-mdaa/bedrock-knowledge-base-l3-construct';

const knowledgeBase = new BedrockKnowledgeBaseL3Construct(this, 'MyKB', {
  kbName: 'my-knowledge-base',
  kbConfig: {
    role: kbExecutionRole,
    vectorStore: 'my-vector-store',
    embeddingModel: 'amazon.titan-embed-text-v2:0',
    s3DataSources: {
      documents: {
        bucketName: 'my-docs-bucket',
        prefix: 'documents/',
        enableMultiSync: true
      }
    }
  },
  vectorStoreConfig: {
    vpcId: 'vpc-12345',
    subnetIds: ['subnet-1', 'subnet-2']
  },
  kmsKey: myKmsKey,
  roleHelper: roleHelper,
  naming: naming
});

Configuration Options

Knowledge Base Properties

role: Execution role for the knowledge base (required)
vectorStore: Reference to vector store configuration (required)
embeddingModel: Embedding model ID or ARN (required)
s3DataSources: Map of S3 data source configurations
vectorFieldSize: Optional vector field size override
supplementalBucketName: Optional supplemental storage bucket

Vector Store Properties

vpcId: VPC where the vector store will be deployed (required)
subnetIds: Subnet IDs for the vector store (required)
port: Optional database port (default: 5432)
engineVersion: Optional Aurora PostgreSQL version
minCapacity: Optional minimum Aurora Capacity Units
maxCapacity: Optional maximum Aurora Capacity Units

S3 Data Source Options

bucketName: S3 bucket name (required)
prefix: Optional S3 prefix filter
enableSync: Enable automatic sync on S3 changes limiting only one file at a time
enableMultiSync: Enable automatic sync on S3 bucket changes processing multiple file changes at the same time
vectorIngestionConfiguration: Advanced parsing and chunking options

Vector Ingestion Configuration

Parsing Strategies:
- BEDROCK_DATA_AUTOMATION: Automated multimodal parsing
- BEDROCK_FOUNDATION_MODEL: Custom foundation model parsing
Chunking Strategies:
- FIXED_SIZE: Fixed token-based chunking
- HIERARCHICAL: Parent-child chunking
- SEMANTIC: Meaning-based chunking
- NONE: No chunking

IAM Permissions

The construct automatically creates IAM policies with permissions for: - Embedding model invocation - Vector database access - S3 data source access - KMS key usage - Data source synchronization - CloudWatch logging

Dependencies

@aws-mdaa/cloudwatch-constructs
@aws-mdaa/dataops-lambda-l3-construct
@aws-mdaa/ec2-constructs
@aws-mdaa/iam-constructs
@aws-mdaa/iam-role-helper
@aws-mdaa/kms-constructs
@aws-mdaa/l3-construct
@aws-mdaa/rds-constructs
@aws-mdaa/construct
aws-cdk-lib
constructs