Benchmarking retrieval for agents on messy real-world company knowledge

A benchmarking study was conducted to evaluate the performance of knowledge retrieval agents on real-world company knowledge, which is often unstructured and disorganized. The study highlights the challenges of this task and proposes a new benchmark dataset, called 'Company Knowledge Bench', to facilitate research and development in this area. The dataset includes information about companies, their products, and their financials, among other details. The study aims to help developers create more effective knowledge retrieval agents that can handle real-world data.