Sensible now recommends queries to streamline onboarding your documents. After ingesting your uploaded document, Sensible uses LLMs to identify extractable data points and auto-generate queries to parse structured data from your document. Sensible’s new Query Group method, which bundles multiple queries together based on their contextual or locational proximity within a document, makes recommended queries a quick and easy way to build a custom parser.
There are three ways to use Sensible’s auto-generated queries:
- Generate an initial query group for new documents
- Add new query groups to existing parsers
- Add new individual queries to existing query groups
Generate an initial query group for new documents
Use recommended query groups to kick off the configuration writing process for newly uploaded documents. In a blank configuration, select “auto-generate fields,” and Sensible will provide ten relevant queries based on the current document page. For the first page, queries are generated from the initial two consecutive pages with meaningful text, bypassing titles or blank pages.
This use case makes it possible to go from a blank state to a publishable configuration in seconds. This is a great way for first-time configuration writers to get a sense of extractable fields within a specific document, as well as learn best practices for writing accurate queries.
Add new recommended query groups to existing configurations
Recommended query groups can also be added to existing configurations. To generate a new query group, select the page you wish to extract from, click “create query group” at the top of the configuration, and then select “Auto-generate” within that card. Sensible incorporates half of each neighboring page to provide additional query generation context. To prevent query duplication, Sensible filters out any repeat queries present in other query groups.
Add new queries to existing query groups
Expand existing query groups by generating recommended queries. Sensible uses the existing queries within the group as context, and then identifies additional queries to extract. You can select one or more of the suggested queries to add to the query group. This process can be repeated until all possible queries are exhausted.
Explore these recommended query group use cases in action in our new demo video:
