2801 - 通过 WatsonX 和 DataStax 进行企业问答并提供 RAG 和反馈.pdf

1、Orlando,FLOctober 69IBM TechXchange 20252801Thomas SchaeckIBM,Distinguished Engineerwatsonx Development Solutions Engineering CoEEnterprise Q&A with RAG andUser Feedback Analyticswith watsonx and DataStaxWhat is Enterprise Q&A with RAG?IBM TechXchange|2025 IBM CorporationCustomers,employees,citizens

2、 have questionson many topics that so far humans needed to answerToday Gen AI can answer most questions onlyToday Gen AI can answer most questions onlyIFIF all required knowledge is made available Must drive to complete and correct corpusWhat does it take?What does it take?-Process&Index Process&Ind

3、ex docs to build knowledge corpus-Answer Questions Answer Questions with high accuracy and low cost-Analyze User feedback Analyze User feedback to know what to improve-ExpertExpert-inin-thethe-looploop to bridge knowledge gapsQ&A with RAG Solutions AcceleratorEnterprise Question Answering over custo

4、mers knowledge base Process and index docs Process and index docs to build hybrid index forcustomer-defined corpus or connect existing index Accurate answers Accurate answers with out-of-the-box parameters,configure/customize IF needed LowLow-cost:cost:Answer re-use reduces GPU cost 10reduces GPU co

5、st 10 1000 x 1000 x when there are many repeat questions User feedback collection and analytics by topicsUser feedback collection and analytics by topicsfor knowledge base owners&stakeholders Actionable drill down Actionable drill down to questions with low rated answers to inform specific improveme

6、nts Expert recommendations Expert recommendations based on expert profile index to bring relevant experts into the loopRuns on watsonx.ai,connects to watsonx.data*or watsonx Discovery*for hybrid indexing and search*watsonx.data Options:Milvus or DataStax *watsonx Discovery technically identical to E