利用 Amazon Bedrock 为 AI 和代理平台提供支持.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.A I M 3 3 0 5Powering AI and agentic platforms with Amazon BedrockLowry SnowPrincipal WW Specialist GenAI Amazon BedrockAWS(He/him)Julia BodeaSenior Product Manager

2、Technical Amazon BedrockAWS(She/her)Dan EricksonDirector of AI EngineeringVercel(He/him)2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaChallenges scaling AI platforms&agentsPillars of an AI+agentic platformModel SwitchingEvaluation ObservabilityCustomer speaker:VercelClosin

3、g 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Top challenges&Pillars of an AI platform 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.“The average organization scraps 46%of their AI proof-of

4、-concept projects prior to production.”Key Findings-Voice of the Enterprise:AI&Machine Learning,Use Cases 2025S&P Global 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.The prototype to production“chasm”Challenges on the path to productionExcitement&potentialMeaningful business va

5、luePOCAI PLATFORM WITH AGENTSPERFORMANCESCALABILITYSECURITYGOVERNANCE 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Top challenges customers reportedAI platforms are built too static and tend to be bespokeRisks associated with exposing non-deterministic LLM outputs to external c

6、ustomersLimited observability across models and pipelinesDifficulty integrating AI into traditional development methods or longstanding business processes 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Simple call to an LLM-AI PlatformSoftware StackLLMAPIAI PlatformModelsAgentsMe