Granted Patent B2

US 12,450,292 · App. 18/601,102 · Granted Oct 21, 2025

Inventors: Fanus Arefaine (San Jose, CA); Michael Paul Bortis (Tustin, CA); Narsingh Rao Chatla (Hyderabad, IN); Austin Lee Grelle (Riverside, IL); Siddharth Jain (Bengaluru, IN); Guangcao Ji (Plano, TX); Anara Myrzabekova (San Francisco, CA); Winthrop Treynor Smith (Littleton, CO); Manesh Saini (New York, NY); Yashjeet Singh (Bengaluru, IN); Ronald Louis Sobey (Porter, TX); Pradyut K. Parida (Hyderabad, IN)

Assignee: Wells Fargo Bank, N.A.

G06F16/90332

Abstract

Claims (41)

1. A method comprising:

receiving a query via a user interface associated with a chatbot system configured to provide information relating to compliance with rules or regulations;

detecting a domain expertise trigger based upon the query using a specialized model feature for generating a response;

embedding text of the query into one or more vectors such that similar words or phrases have closer vector representations;

comparing the one or more vectors from the query with one or more vectors generated from a document;

generating a score for the one or more vectors generated from the document based on a relevance of the one or more vectors generated from the document to the one or more vectors from the query:

selecting a chunk of the document based on the comparing, where the selecting of the chunk of the document includes selecting the chunk including a vector having a highest score;

constructing a prompt to a large language model including the query and the chunk of the document;

routing, based upon the domain expertise trigger, the query to a separate server device storing a domain-specific model of the large language model;

receiving the response to the prompt from the large language model, wherein the response includes an answer to the query and a relevant section of the document; and

presenting the response on the user interface along with the chunk of the document, including:

presenting an icon representing the document on the user interface adjacent to the response; and

displaying text of the chunk upon receiving selection of the icon.

2. The method of claim 1 , further comprising receiving an upload of the document along with the query.

3. The method of claim 2 , further comprising:

parsing the document into text;

splitting the text into chunks; and

embedding the chunks into the one or more vectors generated from the document.

4. The method of claim 1 , further comprising retrieving the document from a database of documents.

5. A system comprising:

a processor; and

a computer-readable medium storing instructions that, when executed by the processor,

cause the system to:

receive a query via a user interface associated with a chatbot system configured to provide information relating to compliance with rules or regulations;

detect a domain expertise trigger based upon the query using a specialized model feature for generating a response;

embed text of the query into one or more vectors such that similar words or phrases have closer vector representations;

compare the one or more vectors from the query with one or more vectors generated from a document;

generate a score for the one or more vectors generated from the document based on a relevance of the one or more vectors generated from the document to the one or more vectors from the query;

select a chunk of the document based on the comparison, where selection of the chunk of the document includes selecting the chunk including a vector having a highest score;

construct a prompt to a large language model including the query and the chunk of the document;

route, based upon the domain expertise trigger, the query to a separate server device storing a domain-specific model of the large language model;

receive the response to the prompt from the large language model, wherein the response includes an answer to the query and a relevant section of the document; and

present the response on the user interface along with the chunk of the document, including to:

present an icon representing the document on the user interface adjacent to the response; and

display text of the chunk upon receiving selection of the icon.

6. The system of claim 5 , comprising further instructions that, when executed by the processor, cause the system to be configured to receive an upload of the document along with the query.

7. The system of claim 6 , comprising further instruction that, when executed by the processor, cause the system to:

parse the document into text;

split the text into chunks; and

embed the chunks into the one or more vectors generated from the document.

8. The system of claim 7 , comprising further instructions that, when executed by the processor, cause the system to retrieve the document from a database of documents.

Assignments (2)

Continuity (1)

References Cited (17)