Table Data Understanding

Vectara now enables you to query the data stored in the tables of your PDFs.

Tables are everywhere—in reports, research papers, invoices, clinical trials, and much more. Yet, AI agents and assistants have had trouble extracting meaningful data from tables embedded in PDFs due to complex structures and inconsistent formatting. We’re excited about this change to our platform because it unlocks more use cases and increased productivity for you and your users.

Key capabilities

Query data from any cell: When considering specific data within a document or table, you can ask questions specific to that information. For example, given a table with columns for 'Low end mortgage rate' and 'High end mortgage rate', you could ask: “What are the ranges of mortgage rates?”
Semantic comparisons: Compare data across cells based on meaning, helping you identify patterns and trends without diving into manual analysis. For example, given a table on response rates in a clinical trial for heart medicine, broken down by symptom, you can ask: “How many participants experienced adverse skeletal effects?”

Why it matters & how it works

Tables are effective tools for communicating large amounts of structured data -- the more data, the better! Ironically, the more data a table has, the more difficult it can be to understand it all and access all of that value. Vectara’s new tabular data understanding capability bridges the gap, empowering users across industries to extract more value from even more data.

For example, each search result based on tabular data will provide a Table view to show the specific table and row where the information was retrieved.

To start using this feature you’ll need to connect with us to enable the feature flag on your account. Today, it is available as an addon for our Enterprise tier customers. Once enabled:

Enable tabular data ingestion for the given file. If using the API, set the table_extraction_config parameter to true for only the requests that contain PDF documents with tables (docs)

If using the Console, enable the toggle on the Data > File uploader page before importing PDF documents that contain tables (docs)

Start querying your data the same way you normally would (e.g. “How much revenue did Dunder Mifflin generate in 2015?”).

View your results, which will now also include a table object in the API response if the result is derived from a table.

There is also a Type column in the console in the results list that indicates where the data was found and provide a view of the relevant table where applicable.

Example use cases

There are countless use cases for extracting data from tables that exist within documents and querying against that data to build or improve understanding. Here’s a few examples (not an exhaustive list):

Healthcare: Extract patient data, treatment outcomes, and clinical trial results directly from PDF reports.
Research: Query data tables from academic papers or experiment results without manual parsing.
Manufacturing and Logistics: Retrieve inventory metrics, shipping details, or production timelines from structured reports.
Finance and Business Analytics: Analyze balance sheets, revenue data, and operational metrics seamlessly.

Conclusion

Table Data Understanding is designed to work across industries and use cases. Enable the feature today and start extracting actionable insights from tables wherever they exist.

To read more about this new feature and learn how to start using it, check out our documentation.

As always, we’d love to hear your feedback! Connect with us on our forums, on our Discord or on our community. If you’d like to see what Vectara can offer you for retrieval augmented generation on your application or website, sign up for an account!