Bifurcated Code Summary Architecture for Cross-Language Clustering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing artificial intelligence solutions for organizing and assembling native code scripts face challenges due to the lack of preexisting training data, difficulty in verifying results, and the transformation of script code to natural language summaries often results in pseudocode specific to semantic structures and programming languages, leading to redundant work and inefficiencies.

Innovation Solution

A bifurcated model architecture comprising a first model to generate code summaries and a second model to compare and categorize them into clusters, using a large language model (LLM) for pre-processing and an NLP model for clustering, ensuring human-readable summaries that are not specific to semantic structures or code formatting.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a single NLP model is used to transform native script code to natural language summaries, then the process is simple, but the code summaries become specific to semantic structures and programming languages, resulting in redundant work

Engineering Contradiction:

Improvemodel architecture complexityVSAvoidcode development efficiency

Core Design Contradiction:

Device complexityVSProductivity

Solution Approach 1:

The patent divides the code processing task into two distinct models: a first model that generates code summaries and a second model that compares and categorizes them. This segmentation prevents the single-model approach from producing language-specific pseudocode summaries, thereby reducing redundant code development while maintaining architectural simplicity through clear functional separation.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The first model acts as an intermediary between the native code and the second NLP model. It transforms code into a standardized summary format that is independent of specific programming languages and semantic structures, enabling the second model to perform effective comparison and categorization without being influenced by code-specific formatting issues.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Extent of automation

If artificial intelligence models are used to organize code scripts, then automation is improved, but the lack of preexisting training data and verification methods reduces reliability

Engineering Contradiction:

Improvecode organization automationVSAvoidresult verification capability

Core Design Contradiction:

Extent of automationVSReliability

Solution Approach 1:

The second model provides feedback by comparing code summaries against existing clusters and categorizing them. This comparison mechanism creates a verification loop that enhances reliability, as the system can identify patterns, detect duplicates, and validate that new code summaries are properly integrated into the existing codebase structure.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The system performs preliminary actions by pre-processing code through the first model to generate standardized summaries before they reach the second model for comparison. This preliminary transformation ensures that all code is processed through a consistent framework, enabling reliable automation even without extensive preexisting training data.

Inventive Principle:

Principle #10Preliminary action

3Manufacturing precision

If code summaries are generated specific to programming languages and semantic structures, then code accuracy is maintained, but adaptability across different codebases is reduced

Engineering Contradiction:

Improvecode summary accuracyVSAvoidcross-language code comparison capability

Core Design Contradiction:

Manufacturing precisionVSAdaptability or versatility

Solution Approach 1:

The first model maintains local quality by preserving the essential logic and functionality of the original code in its summaries, while the second model applies global standardization to enable cross-language comparison. This two-stage approach ensures that code summaries remain accurate to their source while being adaptable to different programming languages and semantic structures.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The system changes parameters by transforming code summaries from a language-specific representation to a standardized format that is independent of specific programming languages. This parameter transformation enables the second model to compare and categorize code across different languages while maintaining the essential accuracy of the original code functionality.

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12541363B2Systems and methods for a bifurcated model architecture for generating concise, natural language summaries of blocks of code

Publication Date: 2026.02.03 CAPITAL ONE SERVICES LLC

AI summary

Systems and methods are described for generating code summaries using bifurcated model architectures. The bifurcated model architecture may comprise a first model that generates code summaries based on native code script, and the second model may compare, de-duplicate, and/or categorize the code summaries into clusters that perform the same or similar functions. That is, the systems and methods may comprise a first model (e.g., a large language model) to predict new content (e.g., a code summary for a given native script code). The outputs of the first model are then inputted into a natural language processing (NLP) model to compare the code summaries. For example, the first model of the bifurcated architecture performs a crucial pre-processing step that prevents the NLP model from generating cluster descriptions that are specific to semantic structures, programming languages, and/or code formatting.