Granted Patent B2

US 12,675,512 · App. 19/058,178 · Granted Jul 7, 2026

Inventors: Javed Qadrud-Din (Union City, CA); Pablo Arredondo (Palo Alto, CA); Walter DeFoor (Rockville, MD); Alan deLevie (Washington, DC)

Assignee: Casetext, Inc.

G06F16/3329G06F40/289G06F40/205

Abstract

Claims (51)

1 . A method comprising:

receiving via a communication interface a request to update a database system based on a natural language input document;

determining an input metadata extraction prompt based on the natural language input document and a prompt template, the prompt template including a fillable portion, the input metadata extraction prompt being determined at least in part by filling the fillable portion with natural language input text selected from the natural language input document, the prompt template and the input metadata extraction prompt each including one or more natural language instructions to generate novel text identifying one or more data values based on the natural language input text, the one or more data values characterizing the natural language input document, the one or more data values corresponding to one or more database fields in the database system;

determining a completed metadata extraction prompt by executing the one or more natural language instructions via a generative language model to generate the novel text;

determining a database query based on the novel text, the database query updating the database system to include one or more database field values determined based on the one or more data values;

executing the database query to update the database system to include the one or more database field values;

receiving from a client machine a document analysis request to analyze a comparison document based on information stored in the database system;

analyzing the comparison document to determine one or more comparison values corresponding to the one or more database field values;

determining a comparison of the one or more comparison values to information retrieved from the database system; and

transmitting a query response message to the client machine based on the comparison.

2 . The method recited in claim 1 , wherein the input metadata extraction prompt includes a natural language description of the one or more database fields, and wherein the one or more data values are generated based on the natural language description.

3 . The method recited in claim 1 , wherein the natural language input document includes a plurality of clauses, and wherein the natural language input text includes a subset of the plurality of clauses.

4 . The method recited in claim 1 , wherein the one or more natural language instructions includes a text formatting instruction to format the novel text based on one or more formatting parameters.

5 . The method recited in claim 4 , wherein the one or more formatting parameters are specified in accordance with JavaScript Object Notation (JSON).

6 . The method recited in claim 1 , wherein the natural language input document is a contract, and wherein a database field of the one or more database fields identifies information related to an exchange value in the contract.

7 . The method recited in claim 1 , wherein the natural language input document is a contract, and wherein a database field of the one or more database fields identifies whether the contract includes an indemnification clause.

8 . The method recited in claim 1 , wherein the one or more data values include one or more clause-level data values, and wherein the one or more database field values include one or more one or more document-level database field values, and wherein determining the database query involves aggregating the one or more clause-level data values to determine the one or more document-level database field values.

9 . The method recited in claim 1 , the method further comprising:

dividing the natural language input document into a plurality of clauses, the input metadata extraction prompt including a question, the one or more natural language instructions including an instruction to the generative language model to identify one or more of the plurality of clauses relevant to answering the question.

10 . The method recited in claim 1 , the one or more natural language instructions including an instruction to the generative language model to include in the novel text an explanation of reasoning employed by the generative language model in determining the one or more data values.

11 . The method recited in claim 1 , the method further comprising:

identifying a plurality of search terms based on a natural language input text query element received from a client machine;

searching the database system using the plurality of search terms to identify a set of search results satisfying the natural language input text query element; and

transmitting information selected from one or more of the search results to the client machine.

12 . One or more non-transitory computer readable media having instructions stored thereon for performing a method, the method comprising:

receiving via a communication interface a request to update a database system based on a natural language input document;

determining an input metadata extraction prompt based on the natural language input document and a prompt template, the prompt template including a fillable portion, the input metadata extraction prompt being determined at least in part by filling the fillable portion with natural language input text selected from the natural language input document, the prompt template and the input metadata extraction prompt each including one or more natural language instructions to generate novel text identifying one or more data values based on the natural language input text, the one or more data values characterizing the natural language input document, the one or more data values corresponding to one or more database fields in the database system;

determining a completed metadata extraction prompt by executing the one or more natural language instructions via a generative language model to generate the novel text;

determining a database query based on the novel text, the database query updating the database system to include one or more database field values determined based on the one or more data values;

executing the database query to update the database system to include the one or more database field values;

receiving from a client machine a document analysis request to analyze a comparison document based on information stored in the database system;

analyzing the comparison document to determine one or more comparison values corresponding to the one or more database field values;

determining a comparison of the one or more comparison values to information retrieved from the database system; and

transmitting a query response message to the client machine based on the comparison.

13 . The one or more non-transitory computer readable media recited in claim 12 , wherein the input metadata extraction prompt includes a natural language description of the one or more database fields, and wherein the one or more data values are generated based on the natural language description.

14 . The one or more non-transitory computer readable media recited in claim 12 , wherein the one or more natural language instructions include a text formatting instruction to format the novel text based on one or more formatting parameters specified in accordance with JavaScript Object Notation (JSON).

15 . The one or more non-transitory computer readable media recited in claim 12 , wherein the natural language input document is a contract, and wherein a database field of the one or more database fields identifies information related to an exchange value in the contract.

16 . The one or more non-transitory computer readable media recited in claim 12 , wherein the one or more data values include one or more clause-level data values, and wherein the one or more database field values include one or more one or more document-level database field values, and wherein determining the database query involves aggregating the one or more clause-level data values to determine the one or more document-level database field values.

17 . The one or more non-transitory computer readable media recited in claim 12 , the method further comprising:

dividing the natural language input document into a plurality of clauses, the input metadata extraction prompt including a question, the one or more natural language instructions including an instruction to the generative language model to identify one or more of the plurality of clauses relevant to answering the question.

18 . A system configured to perform a method, the method comprising:

receiving via a communication interface a request to update a database system based on a natural language input document;

determining an input metadata extraction prompt based on the natural language input document and a prompt template, the prompt template including a fillable portion, the input metadata extraction prompt being determined at least in part by filling the fillable portion with natural language input text selected from the natural language input document, the prompt template and the input metadata extraction prompt each including one or more natural language instructions to generate novel text identifying one or more data values based on the natural language input text, the one or more data values characterizing the natural language input document, the one or more data values corresponding to one or more database fields in the database system;

determining a completed metadata extraction prompt by executing the one or more natural language instructions via a generative language model to generate the novel text;

determining a database query based on the novel text, the database query updating the database system to include one or more database field values determined based on the one or more data values;

executing the database query to update the database system to include the one or more database field values;

receiving from a client machine a document analysis request to analyze a comparison document based on information stored in the database system;

analyzing the comparison document to determine one or more comparison values corresponding to the one or more database field values;

determining a comparison of the one or more comparison values to information retrieved from the database system; and

transmitting a query response message to the client machine based on the comparison.

19 . The system recited in claim 18 , wherein the input metadata extraction prompt includes a natural language description of the one or more database fields, and wherein the one or more data values are generated based on the natural language description.

Assignments (1)

Continuity (4)

References Cited (170)