2072 - 从文档和图像到智能对话：基于 IBM Granite、Docling 和 RAG 的 AI 聊天机器人.pdf

1、Orlando,FLOctober 69IBM TechXchange 20252072From Docs and Images to Intelligent Conversations:AI Chatbots with IBM Granite,Docling&RAGChat With Your DocumentsMuhammad EzzatInjaz Tech,Ai Practice ManagerAbd El-Moniem DarwishInjaz Tech,CEOAgenda0102030405MotivationTechnology StackSolution Architecture

2、DemoQ&AIBM TechXchange|2025 Injaz TechMotivationIBM TechXchange|2025 Injaz TechDepartments in enterprises rely on various data types to support daily operations.Departments in enterprises rely on various data types to support daily operations.EmailsPDFsScanned PDFsSpreadsheetsHuman Resources Departm

3、ent(HR)Chat LogsAudio ScriptsCustomer Support DepartmentEmailsSpreadsheetsEnterpriseEnterpriseOperations DepartmentScanned PDFsImagesSpreadsheets System Data3MotivationIBM TechXchange|2025 Injaz TechWe need the ability to interact with our data to quickly find answers.4UserQuestionAnswerPDFsSpreadsh

4、eetsEmailsChat LogsScanned PDFsImagesTextual DataTurn various data into unified,usable documents.Break down large documents to unlock clearer insights.Make our data easily searchable and accessible.An intelligent agent that lets us interact with our data seamlessly.MotivationIBM TechXchange|2025 Inj

5、az TechComplicated architecture:Different pipeline for each document type.Complicated architecture:Different pipeline for each document type.PDFsSpreadsheetsEmailsChat LogsScanned PDFsImagesTextual DataPreprocessingOCRPostprocessingPreprocessingPreprocessingExtractorPostprocessingVector Datastore Sp

6、ecialized AgentVector Datastore Specialized AgentVector Datastore Specialized AgentOrchestrator Agent5MotivationIBM TechXchange|2025 Injaz TechComplicated architecture:Drawbacks!Complicated architecture:Drawbacks!preprocessingOCRpostprocessingpreprocessingpreprocessingExtractorpostprocessingVector D