在捷蓝航空构建安全且可扩展的 LLM 框架

©2024 Databricks Inc. —All rights reserved1Building a secure & scalable LLM frameworkSai Ravuru -JetBlue, NavindraYadav -TheomJun 12, 2024©2024 Databricks Inc. —All rights reserved©2024 Databricks Inc. —All rights reserved2Technology Stack©2024 Databricks Inc. —All rights reserved©2024 Databricks Inc. —All rights reservedPillars of a successful LLM stackEfficient and promptingOrganizational knowledge repositoryData and access governance•Training•User feedback•Internet access•Curated data•Accuracy testing•Role based access•Regular audits•Logging•Active Directory integration©2024 Databricks Inc. —All rights reserved©2024 Databricks Inc. —All rights reservedGenerative AI maturity curveExpectationEnterprise documentsAccess governanceAutomation agentsExpectedResponsePrompt templatePrompt templateExpectedResponseData governanceReality©2024 Databricks Inc. —All rights reserved©2024 Databricks Inc. —All rights reservedGenerative AI @ scale©2024 Databricks Inc. —All rights reserved6DsPyto the r