当前位置:首页 > 报告详情

AI 代理的实际应用:利用 Databricks 和非结构化数据按需构建非结构化数据.pdf

上传人: Fl****zo 编号:718630 2025-06-22 22页 55.61MB

1、AI Agentsin ActionStructuring unstructured data on demand with Databricks and UnstructuredChris MaddockHead of Product and EngineeringAgendaWho is Unstructured?Pain points and challenges of unstructured data prep Why custom pipelines fall shortHow Unstructured solves itModel Context ProtocolWhat it

2、is,how Unstructured uses MCPAgentic WorkflowsHow MCP unlocks next-gen data processing workflowsAgentic Data Processing In ActionDemo3Agenda1234Who is Unstructured?+10,000paying customers+30M product downloads+155,000organizations using Unstructured$65Mraised since founding in 20225We help organizati

3、ons transform their complex,unstructured data like PDFs,PPTX,HTML files and more into formats compatible with LLMs so employees can chat with their internal data.Company SnapshotAwards6Your LLMs are only as good as the data theyre trained on-80%of your data is trapped in unstructured file types.123E

4、very file type requires its own bespoke processing pipeline Home grown solutions are too brittle for production It requires a lot of time and expertise to build production grade pipelinesSetting The Stage80%Unstructured Data20%Structured Data7Document CleaningCustom Cleaning ScriptsRegExEmbeddingCus

5、tom CodeCustom CodeCustom CodeFile Transformationpython-docxpython-pptxopenpyxlXlrdCamelotpypandocmsg_parserMarkdownNvidia NIMsVLLMsDownstream ConnectionsCustom CodeUpstream ConnectionsCustom CodeIngestion&Preprocessing 1.0Without Unstructured:The Rats NestChunkingCustom CodeS3&Vector StorageSystems

6、 of RecordLLMSystems of RecordGraph&Vector DatabaseObservability&Retrieval8Unstructured saves you tens of thousands of engineering hours.Enterprise Tech StackEnterprise GenAI ETLIngestion&Preprocessing 2.09Azure Blob Storage SearchBoxConfluenceCouchbaseDatabricks VolumesDiscordDropboxElasticSearchGi

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
本文介绍了Unstructured公司及其产品,他们专注于将非结构化数据(如PDF、PPTX、HTML等)转换为与大型语言模型兼容的格式,以便员工能够与内部数据交流。以下是关键点: 1. **公司背景**:拥有超过10,000个付费客户,产品下载量超过3千万次,自2022年成立以来筹集了6500万美元。 2. **非结构化数据挑战**:约80%的数据被困在非结构化文件类型中,每种文件类型都需要独特的处理流程。 3. **解决方案**:Unstructured提供了一套企业技术栈,通过其“ETL”和“Agentic Workflows”简化并自动化数据处理。 4. **模型上下文协议(MCP)**:这是一个开放标准,允许应用程序以一致的方式向大型语言模型提供上下文,类似于AI的USB-C端口。 5. **Agentic Workflows**:AI代理可以自主发现文档,即时构建工作流,并将结构化输出推送到Databricks。 6. **数据处理演示**:在文章中提到了一个演示,展示了如何通过简单的英语提示来操作Unstructured代理。 7. **未来展望**:AI本地的数据工作流程将使ETL过程变得可发现、适应性强、动态、可组合和灵活,且可重复使用。 文章强调了Unstructured通过简化非结构化数据处理,为AI代理开启下一代数据处理工作流的潜力。
"如何高效转化非结构化数据?" "揭秘Unstructured的MCP黑科技!" "AI代理如何实现自动化ETL?"
客服
商务合作
小程序
服务号
折叠