LLM based methods
LLM-based methods
Extract free text from unstructured documents using large language model (LLM)-based SenseML methods. For example, extract information from legal paragraphs in contracts and leases, or results from research papers.
These methods are low-code alternatives to layout-based methods for structured documents, for example, tax documents or insurance forms.
The following topics describe how to author LLM-based methods using the SenseML editor. For information about authoring LLM-based methods using a visual tool instead of JSON, see Prompt tips.
| Method | Example use case | Notes |
|---|---|---|
| List method | ”For each vehicle in an auto insurance declaration, extract the VIN, model, and year.” | Extracts a list of data out of a document, where you don’t know how the data are represented. |
| NLP Table method | ”For each transaction in a bank statement table, extract the date and amount.” | Extracts a list of data out of a document, where you know they’re in a table. |
| Query Group method | ”When does the policy period end?""What are the last 4 numbers of the account?” | Extracts a single fact or data point. |
| Summarizer computed field method | transform extracted data using LLM prompts | Use this method to transform another method’s output when you can’t use types or other computed field methods. |