在生产中运行 Gen AI.pdf

1、?Running Gen AI in productionDenis DjembiSr Solution AdvisorSofiane ChamiDataScientist23 April 2025C2-Restricted use Use case:Legal professionals face challenges in efficiently accessing,retrieving,and analysing legal documents while ensuring compliance and confidentiality.IntroductionKey Features:A

2、dvanced Legal Search Quickly retrieve relevant legal documents with NLP-powered queries.AI-Powered Insights Summarisation and contextual analysis of legal texts.Confidentiality&Compliance Ensures data security and legal adherence.117C2-Restricted use Description:A single LLM retrieves and generates

3、responses based on retrieved knowledge.Pros:?Simplicity in architecture?Lower computational cost?Faster response timeCons:?Limited adaptability?Struggles with complex,multi-step reasoning?Retrieval may lack contextual refinementDescription:Uses multiple specialised agents collaborating for retrieval

4、,analysis,and generation.Pros:?Better handling of complex queries?Enhanced reasoning through multiple perspectives?More robust knowledge synthesisCons:?Higher computational and infrastructural requirements?Increased latency due to agent coordination?More complex implementationClassic RagRag vs Multi

5、 Agent SystemMulti agent system118C2-Restricted use Exploring the RAG ArchitectureDOC,PPTX,PDFSpliter toolEmbedding modelGenerate answers with retriever resultsAzure GPT 4o-mini Vertorised queryResultsRetrieverGeneratorDatabaseSending queryStore split dataAnswerRetrieval Augmented GenerationC2-Restr

6、icted use The Multi Agent ArchitectureVertorized queryResultsOrchestratorGeneratorDatabaseSending queryStore split dataAnswerGenerate answers with retriever resultsAzure GPT 4o-miniDOC,PPTX,PDFSpliter toolEmbedding model120C2-Restricted use Demo121C2-Restricted use Deployment of VeriLex on AzureAzur