Building Advanced RAG Over Complex Documents构建能够处理复杂文档的高级RAG应用-LlamaIndex Talk (Data + AI Summit 2024)数据AI峰会上分享了如何构建能够处理复杂文档的高级RAG应用，几乎涵盖了目前数据处理、RAG、Agentic RAG 等顶级前沿的理念并开放了PPT和代码.pptx

下载文档

0
0
约1.15万字
约 10页
2024-07-03 发布于北京
举报
版权申诉
保障服务

Building Advanced RAG Over Complex Documents构建能够处理复杂文档的高级RAG应用-LlamaIndex Talk (Data + AI Summit 2024)数据AI峰会上分享了如何构建能够处理复杂文档的高级RAG应用，几乎涵盖了目前数据处理、RAG、Agentic RAG 等顶级前沿的理念并开放了PPT和代码.pptx

1、本文档共10页，可阅读全部内容。
2、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。

1BuildingAdvancedRAGOverComplexDocumentsJerryLiu

June11,2024

BuildingaKnowledgeAssistantRAGOverview:BasicRAGandwhereitgoeswrongImprovingDataQuality:ImproveLLMreasoningovercomplexdataWorkshop:LlamaParseoverComplexDocumentsImprovingQueryComplexity:fromRAGtoagentsWorkshop:LlamaParse-powereddocumentagentWhat’snext?2Agenda

3EnterpriseUseCases

4EnterpriseUseCasesDocumentProcessingTaggingExtractionKnowledgeSearchQA(RAG)ConversationalAgentWorkflowAutomationAgent:…Human:…Agent:…DocumentTopic:Summary:Author:KnowledgeBaseAnswer:Sources:…Workflow:ReadlatestmessagesfromuserASendemailsuggestingnext-stepsInboxreadEmailwrite

5BuildingaKnowledgeAssistantAgent:…Human:…Agent:…KnowledgeBaseAnswer:Sources:…Human:QuestionHuman:…

6BuildingaKnowledgeAssistantAgent:…Human:…Agent:…KnowledgeBaseAnswer:Sources:…Human:QuestionHuman:…Goal:Buildaninterfacethatcantakeinanytaskasinputandgivebackanoutput.Inputforms:simplequestions,complexquestions,researchtasksOutputforms:shortanswer,structuredoutput,researchreport

7RAG

8RetrievalAugmentedGeneration(RAG)AnoverviewofaRAGPipelineDataParsingIngestionDataQueryingIndexDataDataParsing+IngestionRetrievalLLM+PromptsResponse

9NaiveRAGPyPDFSentenceSplittingChunkSize256SimpleQAPromptDenseRetrievalTop-k=5IndexDataDataParsing+IngestionRetrievalLLM+PromptsResponse

10ChallengeswithNaiveRAG

NaiveRAGapproachestendtoworkwellforsimplequestionsoverasimple,smallsetofdocuments.“WhatarethemainriskfactorsforTesla?”(overTesla202110K)“WhatdidtheauthordoduringhistimeatYC?”(PaulGrahamessay)11EasytoPrototype,HardtoProductionize

12EasytoPrototype,HardtoProductionizeButproductionizingRAGovermorequestionsandalargersetofdataishard!

FailureModes:SimpleQuestionsoverComplexDataSimpleQuestionsoverMultipleDocumentsComplexQuestions13EasytoPrototype,Hardt