New serverless pattern - apigw-http-api-lambda-bedrock-dynamodb-vector-search-cdk

[proton0210]

Summary

• add a TypeScript AWS CDK pattern for API Gateway HTTP API, Lambda, Amazon Bedrock embeddings, and native DynamoDB vector search
• add POST /documents for embedding and storing documents and POST /search for filtered semantic retrieval with SearchVectors
• use a tenant-scoped composite primary key (tenantId, documentId) so the same document ID can safely exist in different tenants
• validate required fields, DynamoDB key byte limits, Titan's 50,000-character input limit, metadata lengths, and topK before invoking AWS services
• add a bounded CloudFormation custom resource that creates, waits for, replaces, and deletes the DynamoDB vector index while native CDK/CloudFormation coverage is unavailable
• manage all application and custom-resource Lambda log groups in CloudFormation with one-week retention and DESTROY, preventing cleanup leftovers
• add least-privilege IAM, sample HTTP and Lambda events, complete deployment/testing/cleanup documentation, optional credential-free CI/CD deployment steps, and Serverless Land metadata

Why this pattern is useful

This pattern gives serverless developers a focused starting point for semantic search while keeping embeddings beside operational data in DynamoDB. It is useful because it:

• removes the need to provision a separate vector database or maintain a change-data-capture synchronization pipeline
• demonstrates tenant-scoped vector partitions and optional category filters for SaaS, document discovery, product catalog, support, and knowledge-search APIs
• bridges the current CDK and CloudFormation coverage gap with an idempotent, bounded custom resource for DynamoDB vector-index lifecycle management
• combines a pay-per-request HTTP API, Lambda compute, Bedrock embeddings, least-privilege IAM, sample events, and deployment tests in a small reusable pattern rather than a full application

The collection currently contains an S3 Vectors RAG example but no native DynamoDB vector-search pattern.

Validation

• clean Node.js 24 dependency installation, npm run build, npm test, and npm run synth -- --quiet passed
• all 9 tests passed across 3 suites, covering application behavior, request boundaries, vector-index lifecycle behavior, and synthesized infrastructure
• repository example-pattern.json schema validation passed; its introduction is 354 words and it contains 4 official resources
• generated template inspection found 22 resources, all 3 Lambda log groups managed by CloudFormation, no Step Functions resource, and zero wildcard IAM actions or resources
• exact commit cf375c1adc3f2a81e9a330163604bdf5bbc92602 deployed through the documented interactive npx cdk deploy path to CREATE_COMPLETE
• both documented HTTP ingests returned 201; semantic search returned both records on the first attempt and ranked the DynamoDB vector-search document first
• both supplied direct-Lambda events completed with Lambda status 200 and application responses 201 and 200
• live negative tests returned 400 for oversized model input, a multibyte oversized tenant key, and malformed JSON; an unknown route returned 404
• the same documentId was stored in a second tenant and its filtered search returned exactly one tenant/category-correct result with zero cross-category matches
• application and provider logs contained 38 events with zero ERROR, Exception, Request failed, or timeout matches
• the documented npx cdk destroy path removed the stack; the README confirmation returned [], and post-destroy inventory found zero residual tables, Lambda functions, HTTP APIs, IAM roles, or log groups belonging to the pattern

Notes

• Serverless Land submission issue: New pattern submission - apigw-http-api-lambda-bedrock-dynamodb-vector-search-cdk #3276
• The pattern uses an unauthenticated HTTP API for concise demonstration and documents production authorization/CORS considerations.
• The vector-index provider uses a bounded 13-minute wait intended for the new empty table deployed by this pattern.
• No account-specific outputs, API endpoints, role ARNs, AWS profiles, credentials, or generated deployment files are included in this PR.
• npm audit --omit=dev currently reports one high-severity transitive brace-expansion advisory bundled inside aws-cdk-lib@2.263.0; npm cannot replace that bundled dependency automatically, and it is CDK tooling rather than deployed Lambda code.