Just-in-Time Knowledge Base with Amazon Bedrock

Amazon Bedrock empowers SaaS companies to build cost-effective, multi-tenant knowledge bases. This innovative solution addresses the challenges of managing vast document collections for numerous clients with varying needs and scales resources based on actual usage. Unlike traditional Retrieval Augmented Generation (RAG) systems, this just-in-time approach processes documents only when needed, eliminating unnecessary storage and processing costs associated with unused data. The solution leverages serverless technologies like Amazon Bedrock‘s CUSTOM data source type, Amazon OpenSearch Serverless for efficient querying, and Amazon DynamoDB with Time-to-Live (TTL) for automated document expiration. This ensures a lean system focused on relevant content while maintaining strict data isolation between tenants. Multi-tenant isolation is achieved through metadata-based filtering at the user, tenant, and file levels, enabling granular control and configurable resource limits (ingestion rates, query rates, TTL durations). Tiered pricing models are supported, allowing SaaS providers to offer various service levels based on document storage, query limits, and TTL settings. The architecture uses pre-signed URLs for secure file uploads to Amazon S3, DynamoDB for tracking documents with TTL, and Lambda functions for automated cleanup of expired documents. The system also allows for contextual querying, returning both AI-generated answers and source citations. While offering significant cost advantages and scalability, potential drawbacks could include complexity in initial setup and the need for expertise in AWS services. Compared to traditional RAG systems, this solution offers superior cost efficiency and scalability, making it ideal for SaaS companies managing a large number of small to mid-sized tenants with potentially transitory projects.

(Source: https://aws.amazon.com/blogs/machine-learning/build-a-just-in-time-knowledge-base-with-amazon-bedrock/)