W4_Sulin(1).pdf

1、Hands-on RAG WorkshopJeff FriedPatrick SulinSuprateem BanerjeeInterSystemsInterSystems Corporation.All Rights Reserved.Over 40%of agentic AI projects will be canceled by the end of 202752%of generative AI(GenAI)projects will be retired after the proof-of-concept(POC)stage through 2026Results demonst

2、rate that RAG is essential:Without retrieval,exact match accuracy is 0%across all tasks,whereas retrieval yields substantial gains in execution accuracy(up to 79.30%)and component match accuracy(up to 78.86%)arXiv:2602.07086 cs.SE 6 Feb 2026AgendaIntroductionsRAG Why and WhatFramework(7 steps to suc

3、cessful RAG)RAG Hands-on TutorialRAG RecipesAgentic AI&Context EngineeringResearch&Industry DirectionsTakeawaysYour Instructors TodayJeff FriedPatrick SulinSuprateem BanerjeeFounding SponsorContact coordinates:Handle 60%of NYStock Exchange trafficManage over 1 BillionPatient Records WorldwideTrack 2

4、0 Million Shipping ContainersInterSystems Impact in Healthcare,Financial Services,and Supply ChainUnparalleled performance,scalability,interoperability,and reliabilityHands-on -primary resourcesDO Why and WhatFramework(7 steps to successful RAG)RAG Hands-on TutorialRAG RecipesAgentic AI&Context Engi

5、neeringResearch&Industry DirectionsTakeawaysThe core idea of RAGA model by itself=static knowledge+pattern prediction RAG=model+real-time information retrieval That combination changes a lot.1.Keeps answers up to dateModels have a knowledge cutoff.RAG lets them pull in current data(docs,databases,th

6、e web),so answers dont get stuck in the past.2.Reduces hallucinationsWithout grounding,models may confidently invent details.RAG forces the model to base its response on retrieved sources,which makes answers more factual and traceable.3.Makes responses domain-specificA base model doesnt know your co