JSON Parsing Layer for Accurate LLM Analysis of Quality Records

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for analyzing quality records, which are stored as JSON objects and attached files, are inefficient and inaccurate due to the language-independent nature of JSON objects and the requirement for clear English context by large language models (LLMs).

Innovation Solution

A system and method that converts JSON objects and attached files into English language-based textual data using APIs and applicable rules, enabling accurate analysis and summary generation through large language models.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If JSON objects are used to store quality records, then data structure and storage efficiency are improved, but interpretability and analysis accuracy deteriorate because JSON is not in natural language

Engineering Contradiction:
Improvedata storage efficiencyVSAvoiddata interpretability
Core Design Contradiction:
ProductivityVSEase of operation

Solution Approach 1:

The patent introduces an intermediary translation layer that converts JSON objects into natural language text. This mediator component receives structured JSON data, applies language models to translate it into human-readable format, and outputs contextualized text that preserves the original data structure while adding interpretability. This resolves the contradiction by maintaining JSON storage efficiency while enabling natural language analysis.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Extent of automation

If LLMs are used to analyze quality records, then automation and efficiency are improved, but accuracy deteriorates because LLMs require clear English context which JSON does not provide

Engineering Contradiction:
Improveanalysis automationVSAvoidanalysis accuracy
Core Design Contradiction:
Extent of automationVSMeasurement precision

Solution Approach 1:

The patent applies preliminary action by translating JSON objects into natural language text before feeding them to the LLM. This preprocessing step ensures that the LLM receives input in the required format with proper context and language structure, thereby maintaining high automation while improving analysis accuracy. The translation occurs in advance, preparing the data optimally for LLM processing.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If manual analysis of quality records is performed, then accuracy can be maintained through human judgment, but productivity and efficiency deteriorate due to the difficulty of interpreting JSON files

Engineering Contradiction:
Improveanalysis accuracyVSAvoidanalysis efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent implements self-service by enabling the system to automatically translate and analyze quality records without requiring manual human intervention. The translation layer autonomously converts JSON to natural language, and the LLM automatically performs the analysis, eliminating the need for manual JSON interpretation while maintaining high accuracy. This resolves the contradiction by achieving both automation and precision.

Inventive Principle:
Principle #25Self-service

Data Source

PatentEP4679289A1System and method for handling of JSON objects by a large language model
Publication Date: 2026.01.14 HONEYWELL INTERNATIONAL INC
  • EP4679289A1 patent drawingFigure 1
  • EP4679289A1 patent drawingFigure 2
  • EP4679289A1 patent drawingFigure 3

AI summary

The present disclosure discloses a system and method for handling of JSON objects by a large language model (LLM). The proposed system and method involves receiving a user input requesting an analysis of one or more quality records, and extracting the one or more of quality records, wherein each quality record comprises one or more JSON objects. The system and method further involve identifying an applicable rule for parsing each of the one or more JSON objects, parsing each of the one or more JSON objects by an API corresponding to the identified applicable rule, feeding the parsed JSON objects to the LLM to obtain the analysis of the one or more quality records, and displaying the obtained analysis of the one or more quality records to the user.