从试点到生产：金融服务的LLMOps.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.I N D 3 7 6From pilots to production:LLMOps for financial servicesArnav KharePrincipal Solutions ArchitectGlobal Financial ServicesAmazon Web ServicesJorge Castans G

2、arciaSr.Solutions ArchitectGlobal Financial ServicesAmazon Web Services 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaIntroductionChallenges in GenAI and how LLMOps can helpMLOps/LLMOps for fine-tuning modelsGenAI Ops for building AI applicationsAgent Ops for building Agen

3、tic applications 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AIOps TaxonomyAI GovernanceAI GovernanceMLOpsMLOpsGenAIOpsGenAIOpsAIOpsForms the foundation for managing infrastructure,data,models,and AI applications throughout their lifecycles.Builds on AI Governance and incorpor

4、ates DevOps principles to streamline the build,deployment,and monitoring of traditional ML models.Extends MLOps to support GenAI applications,utilizing foundation models and powering Retrieval Augmented Generation(RAG)and Agentic workflows.2025,Amazon Web Services,Inc.or its affiliates.All rights re

5、served.How organizations measure success for AI/ML projects1.Success rateQualifyInitiateBuildValidateDeployScale2.Time-to-value3.Quality/Defect rateJourney of an AI/ML use case from ideation to production 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Many organizations struggle

6、with these metrics4630%Fail to make to production1GenAI projects abandoned211mo.Time-to-value31.Source:Gartner AI in Organizations Survey,20222.Source:Press Release on Gartner Data&Analytics Summit,July 20243.Based on the average time of customers interviewed 2025,Amazon Web Services,Inc.or its affi