EnterpriseRAG-Bench
A dataset and benchmark of more than 500,000 simulated company documents and 500 questions for evaluating retrieval-augmented generation systems on realistic enterprise data.
EnterpriseRAG-Bench provides a dataset simulating a fictional AI-inference company, with slightly over 500,000 internal documents and 500 questions, released to help teams benchmark retrieval-augmented generation systems and fine-tune agents on realistic enterprise data. The code additionally generates similar datasets for different industries and company scales.
Its documents are drawn from sources such as Slack channel discussions, Gmail email threads, Linear project-management tickets, Google Drive files, Hubspot CRM records, Fireflies meeting transcripts, GitHub pull requests and comments, Jira tickets, and Confluence wikis and documentation, covering planning, engineering, sales, customer success and internal communication.
Its 500 questions are split into ten categories of increasing difficulty, from simple single-document questions and semantically rephrased questions with little keyword overlap, through questions that require combining information across sections of one document or across several related documents, questions with disqualifying constraints, and questions where source documents directly contradict each other.