Welcome! Sensible is a developer-first platform for extracting structured data from documents, for example, business forms in PDF format. It's highly configurable: you can get simple data in minutes by leveraging GPT-4 and other large-language models (LLMs), or you can tackle complex and idiosyncratic document formatting with Sensible's powerful layout-based document primitives.
See the following list for an overview of going live with Sensible:
- Learn to extract data, or use out-of-the-box supported document types
- Integrate using Sensible's API, quick-extract UI, or other tools
- Quality control extracted data
- Monitor extracted data in production
This guide gets you started with the first step, extracting data.
Let's get started with extracting document data from an example bank statement. We'll use a large-language model (LLM) prompt to extract a checking account number in minutes.
In this guide, you'll:
- Extract data from an example document using a natural-language description of your target data
- Publish your prompt as part of a "config."
- Test your config against a second, similar document to ensure it extracts the same target data.
Get an account at sensible.so. If you don't have an account, you can still read along to get a rough idea of how things work.
Log into the Sensible app.
To view an example bank statement extraction, navigate to https://app.sensible.so/editor/instruct/?d=sensible_instruct_basics&c=bank_statement&g=bank_statement.
Sensible displays an example document in the left pane, and fields of extracted data in the right pane.
Take the following steps to create a prompt to extract more data from the document.
- To extract a single data point from the document, click Query.
- Edit the query as shown in the following screenshot by entering
checking account number (not savings)in the query field. Give the query an ID,
account_num_checking, then click the Send icon:
- Sensible displays the extracted account number,
8347-32348, in the Extracted data section.
- Click Back to fields.
Congratulations! You extracted the checking account number from the bank statement.
To extract checking account numbers from other bank statements, you'll need to publish the "config" containing your prompt.
Click Publish, click Production, then click Publish to production:
Let's see if the config containing your prompt works with other bank statements. To test the prompt, take the following steps:
- Navigate to https://app.sensible.so/editor/instruct/?d=sensible_instruct_basics&c=bank_statement&g=bank_statement_2. Notice that the left pane now displays a statement for a different customer.
- In the right pane, scroll down to the
account_num_checkingfield you authored in previous steps. Verify that the extracted information automatically updated to reflect the second example document. For example, the account number updated from
It looks like your prompt was successful at extracting the checking account number from another document. Great!
Try extracting other pieces of information using what you learned in previous steps, such as:
- The customer's address
- The Spanish-speaking customer service phone number
- The time period for each account. Hint: Use the List method. For example, in this config, the
accounts_listuses the List method.
Publish the config to save your changes.
To extract data from your documents, first check if they're on Sensible's list of out-of-the-box supported document types. If not, configure your custom extractions by using the interactive tutorial or taking the following steps:
- To exit the Sensible Instruct editor, click Sensible in the upper left corner.
- Click the Document types tab. Create a new document type, then click the type you created to edit it.
- In the document type's Reference documents tab, upload your own example document.
- In the document type's Configurations tab, create a new test configuration, and click the configuration you created to edit it.
- Write prompts in the configuration editor to extract data using what you learned in previous steps.
Explore extracting lists, tables, and single data points with other interactive examples:
For advanced extraction strategies, see Choosing an extraction approach
Get extracted document data out of Sensible and put it to work in Excel files, databases, and other destinations. See Integrating.
Updated about 6 hours ago