Insurance resolution platform

A system for automating a comparison of documents, comprising:

• a front-end comprising a user interface, the user interface configured to provide interactive tools for a user to review, filter, and modify the one or more documents;
• a back-end communicating with the front-end via one or more application program interfaces (APIs), the back-end configured to execute one or more tasks asynchronously and concurrently using an asynchronous framework for non-blocking operation, one or more machine learning models, the back-end comprising:
• an input module configured to receive one or more documents in varying formats and preprocess the one or more documents by, converting the one or more documents into a machine-readable text using machine learning-based optical character recognition, and analyzing the layouts using pattern recognition algorithms and learned visual segmentation models in the one or more documents;
• a data extraction module employing contextual natural language processing algorithms to extract one or more predetermined data items from the machine-readable text;
• and
• a data normalization module configured to:
• serialize the extracted data into structured JSON objects, wherein the JSON objects comprise key-value pairs;
• apply schema mapping to align the one or more predetermined data items to a predefined key-value structure;
• and
• generate semantic embeddings for description fields and compute cosine similarity to resolve terminological inconsistencies;
• a comparison module configured to:
• compute cosine similarity scores between embeddings of description fields;
• determine matched pairs exceeding a configurable similarity threshold;
• exclude matched pairs from subsequent comparison iterations to prevent duplication by adding matched pairs to a tracking data structure;
• identify unmatched items;
• and
• generate structured matched and unmatched output as a consolidated JSON data structure containing matched pairs with similarity scores and unmatched items with placeholder indicators;
• and
• a machine learning component configured to update one or more of the input module, data extraction module, data normalization module, and comparison module using user-provided feedback.

A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform a method for comparing documents, the method comprising:

• receiving, via a user interface on a front-end computing device, one or more documents in one or more formats;
• sending the documents to a back-end computing system configured for non-blocking operation using an asynchronous framework to execute tasks concurrently;
• processing the one or more documents using machine learning-based optical character recognition and layout analysis;
• extracting one or more data fields using natural language processing techniques including named entity recognition and contextual parsing;
• serializing the extracted data fields into structured JSON representations, wherein the JSON representations comprise key-value pairs;
• normalizing, the one or more data fields into a predefined format using schema mapping and semantic embedding-based synonym resolution including cosine similarity computation to form a normalized data;
• comparing the normalized data using cosine similarity scores between semantic embeddings;
• selecting matched pairs exceeding a configurable similarity threshold;
• excluding matched pairs from further comparison iterations by adding matched pairs to a tracking data structure;
• identifying unmatched data fields;
• performing the comparison asynchronously across serialized JSON representations;
• outputting a consolidated JSON data structure containing matched pairs with similarity scores and unmatched items with placeholder indicators;
• generating a visual report of the discrepancies;
• and
• adapting the method by training one or more machine learning models in response to using the user interaction as feedback to improve accuracy of the processing, extracting, normalizing, and comparing.