编写具有生存时间 (TTL) 的低成本多租户知识库.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.S A S 4 1 2Coding Cost-Efficient Multi-Tenant Knowledge Bases with TTLSteven WarwickSr.Solutions ArchitectMike GillespiePrincipal Solutions Architect 2025,Amazon Web

2、 Services,Inc.or its affiliates.All rights reserved.AgendaWhy Knowledge Bases?Challenges and solutions for RAG Multi-tenancyDemoQuestions and Answers 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Design Decisions In Knowledge BasesTenant Specific SearchHybrid SearchAI-enabled fe

3、aturesApplication enables users to upload documents and then allow users to search against them.Do full text search and semantic search.Power chatbots or other AI features that are customized with tenant specific data.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web

4、 Services,Inc.or its affiliates.All rights reserved.Challenges and Solutions for Multi-tenant Knowledge Bases 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.ChallengesData isolationHow can a SaaS provider cost-effectively provide tenant data isolation,while still ensuring that da

5、ta never leaks across tenant boundaries.Tiered performanceHow does a SaaS provider enable tiered experience that enables customers to pay more for a higher tier of service.For example,can you make data volume and latency dependent on the subscription level.Data lifecycle managementProvide the capabi

6、lity to have a data expire from the search,or tiered to another level 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Challenge:Tiered PerformanceOpensearchpgVectorS3 Vector BucketLow latencyLargest Vector StoresHybrid SearchProvisioned Cost