在复杂文档上构建生产 RAG.pdf

1、2024 Databricks Inc.All rights reserved1Building Advanced Building Advanced RAG Over Complex RAG Over Complex DocumentsDocumentsJerry LiuJerry LiuJune 11,2024June 11,20242024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved1.1.Building a Knowledge AssistantBuilding a Knowled

2、ge Assistant2.2.RAG Overview:Basic RAG and where it goes wrongRAG Overview:Basic RAG and where it goes wrong3.3.Improving Data Quality:Improving Data Quality:Improve LLM reasoning over complex dataWorkshop:Workshop:LlamaParse over Complex Documents4.4.Improving Query Complexity:Improving Query Compl

3、exity:from RAG to agents Workshop:Workshop:LlamaParse-powered document agent5.5.WhatWhats next?s next?2AgendaAgenda2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved3Enterprise Use Enterprise Use CasesCases2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rig

4、hts reserved4Enterprise Use CasesEnterprise Use CasesDocument ProcessingTagging&ExtractionKnowledge Search&QA(RAG)Conversational AgentWorkflow AutomationAgent:Human:Agent:DocumentTopic:Summary:Author:Knowledge BaseAnswer:Sources:Workflow:Read latest messages from user ASend email suggesting next-ste

5、psInboxreadEmailwrite 2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved5Building a Knowledge AssistantBuilding a Knowledge AssistantAgent:Human:Agent:Knowledge BaseAnswer:Sources:Human:Human:2024 Databricks Inc.All rights reserved2024 Databricks Inc.All rights reserved6B

6、uilding a Knowledge AssistantBuilding a Knowledge AssistantAgent:Human:Agent:Knowledge BaseAnswer:Sources:Human:Human:Goal:Goal:Build an interface that can take in any task as input and give back an output.Input forms:Input forms:simple questions,complex questions,research tasksOutput forms:Output f