在 Databricks 上构建面向用户的 AI 系统的最佳实践.pdf

《在 Databricks 上构建面向用户的 AI 系统的最佳实践.pdf》由会员分享，可在线阅读，更多相关《在 Databricks 上构建面向用户的 AI 系统的最佳实践.pdf（33页珍藏版）》请在三个皮匠报告上搜索。

1、Best Practices forBuilding User-Facing AI Systems on DatabricksArthur DonnerJyotsna BharadwajAbout UsArthur Dooner,Specialist Solutions ArchitectJyotsna Bharadwaj,Sr Solutions ArchitectYouve heard the story already on Building AI Agents,but lets recapAgent that Orchestrates Other Agentsaka“multi-age

2、nt”Common AI System DesignsBatch InferenceAka“AI-Query”Agent with Documents or SQL(Genie)aka unstructured retrieval/RAG and“text-to-SQL”Increasing Functionality,but also ComplexityAgent with multiple tools+data sourcesCompound AI systemaka“tool calling agent”Mosaic AI supports these and moreAgents a

3、nd tools that respect existing data and AI governanceShave months off your time-to-market with our agent toolingMosaic AI Enables you to Build Production-Quality,Enterprise-Ready Agents FasterDeliver accurate agents that are evaluated and monitoredProduction qualityEnd to end governance7Reduce poten

4、tial privacy&reputational riskProvides native evaluation and monitoring,grounded in our bleeding edge researchRapidly iterate and redeploy to improve qualityRapid developmentYour AI System Exists Now How Will Users Experience It?Scaling AI Means Scaling the User ExperienceRapidly prototype AI insigh

5、ts through pre-built frontends and testing tools.Give space for users to provide feedback.Reach your first user set:Simpler frontendsChat&bi platformsAsync insights Fit into their day to day.First stepsPilot valueCombine AI insights across different project streams.Serve powerful frontends that beco

6、me the“hub for AI insights”.Broad adoptionBetter AI Bigger Audiences Evolving UXAI Integrations On The Databricks Platform10BATCH INFERENCEUSE SQL EDITORDEPLOY TO USERS Bring AI To Where Users NeedAI Integrations Outside of Databricks11Chat Applications&MailReporting and Business IntelligenceExterna