Computer-implemented method and system for analysis and validation of consistency across documents including images and language data using artificial intelligence (AI)
A computer-implemented method provides AI-driven quality assurance (QA) on damage claims through consistency checks across claims data at different processing stages and against policy, regulation, and guideline information. An AI-based claims QA agent receives claims data comprising a photo document, a claim document, and a reference document for a damaged property. The agent extracts images and a textual description from the photo document, processes the images using a computer vision model to extract first damage information, and processes the textual description using a machine learning (ML) model to extract second damage information. The agent validates accuracy of the photo document based on consistency between the first and second damage information, combines that information into third damage information, and validates accuracy of the claim document against contextual information from the reference document and the third damage information. A large-language model (LLM) generates a QA report identifying issues, initiating corrective updates.
1 . A computer-implemented method for providing efficient, systematic artificial intelligence (AI)-driven quality assurance (QA) assessment on damage claims based on an analysis and multiple consistency checks across claims data from different processing stages and against relevant policy, regulation, and guideline information, the method comprising:
receiving, by an AI-based claims QA agent including instructions stored in non-transitory memory of a computer system and executable by a processor of the computer system, claims data comprising a photo document, a claim document, and a reference document associated with a damaged property;
extracting, by the AI-based claims QA agent, one or more images and a textual description of the damaged property from the photo document;
processing, by the AI-based claims QA agent, the one or more images using a computer vision model to extract first damage information associated with the damaged property;
processing, by the AI-based claims QA agent, the textual description using a first machine learning (ML) model to extract second damage information associated with the damaged property;
validating, by the AI-based claims QA agent, an accuracy of the photo document based on a consistency between the first damage information and the second damage information to generate a first consistency check result;
combining, by the AI-based claims QA agent, the first damage information and the second damage information into third damage information;
processing, by the AI-based claims QA agent, the claim document using a second ML model to extract claim information associated with the damaged property, wherein the claim information comprises a geographical location of the damaged property and one or more line items associated with restoring the damaged property;
processing, by the AI-based claims QA agent, the reference document using a third ML model to extract contextual information associated with at least one of a policy, a regulation, or a guideline associated with a specific geographical location;
validating, by the AI-based claims QA agent, an accuracy of the claim document based on a consistency of the claim information against the contextual information, and the combined third damage information to generate a second consistency check result;
initiating, by the AI-based claims QA agent, a large-language model (LLM) to generate, based on the first consistency check result, the second consistency check result, and the reference document, a QA report for the claims data;
receiving, by the AI-based claims QA agent, from the LLM, the QA report indicating an issue in at least one of the photo document, the claim document, or the reference document; and
initiating, by the AI-based claims QA agent, based on the QA report, updating the at least one of the photo document, the claim document, or the reference document to correct the issue.
2 . The method of claim 1 , wherein the first damage information extracted from the one or more images of the photo document and the second damage information extracted from the textual description of the photo document are associated with at least one of a damage type, a damaged object, a quantity of damaged objects, or a material type associated with the damaged property.
3 . The method of claim 1 , wherein an individual one of the one or more line items in the claim information extracted from the claim document comprises at least one of a service category, a description, a unit price, or a quantity.
4 . The method of claim 1 , wherein the validating the accuracy of the photo document comprises:
computing, by the AI-based claims QA agent, a similarity measure between the first damage information extracted from the one or more images of the photo document and the second damage information extracted from the textual description of the photo document; and
comparing, by the AI-based claims QA agent, the similarity measure to a threshold.
5 . The method of claim 1 , wherein the validating the accuracy of the claim document comprises:
computing a similarity measure between:
(1) the claim information extracted from the claim document, and
(2) the contextual information extracted from the reference document and the combined third damage information from the photo document; and
comparing, by the AI-based claims QA agent, the similarity measure to a threshold.
6 . The method of claim 1 , wherein the validating the accuracy of the claim document comprises:
computing a similarity measure between previously processed claims data associated with the damaged property and the claim information extracted from the claim document; and
comparing, by the AI-based claims QA agent, the similarity measure to a threshold.
7 . The method of claim 1 , wherein the validating the accuracy of the photo document comprises:
initiating, by the AI-based claims QA agent, a second LLM to generate the first consistency check result based on the first damage information extracted from the one or more images of the photo document, the second damage information extracted from the textual description of the photo document, and the photo document; and
receiving, by the AI-based claims QA agent, from the second LLM, the first consistency check result comprising at least one of:
a textual description of at least one of an approval or a discrepancy between the one or more images and the textual description of the photo document, or
an annotated photo document highlighting a discrepancy.
8 . The method of claim 1 , wherein the validating the accuracy of the claim document comprises:
initiating, by the AI-based claims QA agent, a second LLM to generate the second consistency check result based on the combined third damage information from the photo document, the claim information extracted from the claim document, the contextual information extracted from the reference document, the photo document, the claim document, and the reference document; and
receiving, by the AI-based claims QA agent, from the second LLM, the second consistency check result comprising at least one of:
a textual description of at least one of an approval or a discrepancy among the photo document, the claim document, and the reference document,
an annotated photo document highlighting a first discrepancy,
an annotated reference document highlighting a second discrepancy, or
an annotated claim document highlighting a third discrepancy.
9 . The method of claim 1 , wherein the validating the accuracy of the photo document further comprises:
determining, by the AI-based claims QA agent, whether there is pre-existing damage on the damaged property based on a comparison of the one or more images against a geographical image database.
10 . The method of claim 1 , wherein the validating the accuracy of the claim document further comprises one or more of:
determining, by the AI-based claims QA agent, whether an individual one of the one or more line items is excluded from coverage based on the policy; or
determining, by the AI-based claims QA agent, whether an estimated cost in an individual one of the one or more line items is inaccurate based on the contextual information and a policy list database comprising price codes.
11 . The method of claim 1 , wherein the initiating the LLM to generate the QA report for the claims data comprises:
providing, by the AI-based claims QA agent, to the LLM, the contextual information extracted from the reference document, the first consistency check result, the second consistency check result, and an input prompt to check for a recent policy change.
12 . The method of claim 1 , wherein the computer vision model used for processing the one or more images to extract the first damage information associated with the damaged property comprises at least one of an image segmentation model, a classification model, or a large-vision model (LVM).
13 . The method of claim 1 , wherein at least one of the first ML model, the second ML model, or the third ML model used respectively for processing the textual description of the photo document, the claim document, or the reference document comprises at least one of a name property recognition (NER) model or an LLM.
14 . The method of claim 1 , further comprising one or more of:
preprocessing, by the AI-based claims QA agent, the one or more images of the photo document before processing the one or more images using the computer vision model, wherein the preprocessing comprises at least resolution enhancement; or
preprocessing, by the AI-based claims QA agent, the textual description of the photo document before processing the textual description using the first ML model, wherein the preprocessing comprises at least one of a handwritten recognition enhancement or a contextual enhancement.