Get answers from tables with tabular search in Amazon Bedrock Managed Knowledge Base
Amazon Bedrock Managed Knowledge Base now supports tabular search, so agentic retrieval can answer questions over CSV, XLSX, and XLS files. It looks up values and runs calculations across every relevant row, then combines the results with facts from related documents to answer multi-step analytical questions.
Amazon Bedrock Managed Knowledge Base now supports tabular search, enabling answers to be derived from tables using agentic retrieval. When a question requires data from CSV, XLSX, or XLS files, tabular search retrieves values and performs calculations across all relevant rows in one or more tables or sheets. It then integrates these results with facts from related documents to respond to complex analytical questions that involve multiple steps.
For instance, a finance team needing to know "Which departments must write variance commentary?" can ask tabular search, as the answer depends on several factors, including the variance policy's threshold, budget revisions, and calculations across every department in the profit and loss (P&L) workbook. Unlike retrieval, which returns fixed passages from documents, tabular search splits the work by first retrieving only the relevant passages from the correct table file and then using computational code to process the entire table.
The process begins when a question from a CSV, XLSX, or XLS file is submitted to an AgenticRetrieveStream request, which functions similarly regardless of the connector used (Amazon S3, Microsoft SharePoint, Atlassian Confluence, etc.). After retrieving passages, the agent determines whether to execute code over the retrieved tables.
This computed code runs in an isolated sandbox and returns only the output needed, such as totals or matching rows. The agent then synthesizes this information with the retrieved passages, citing the appropriate sources.
Tabular search is particularly useful for questions requiring calculations or lookups across table files, as it automatically handles the complex relationships within tables that retrieval struggles with. However, it does add an extra agent turn due to the computational step, resulting in longer response times compared to purely text-based answers.
For businesses managing both narrative documents and data tables, tabular search offers a comprehensive solution, combining the strengths of both retrieval and computation without the need for predefined schemas.
Written by urgent.news from AWS Machine Learning's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.