Schema-Guided LLM API Calls for Large Data Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Large Language Models (LLMs) face limitations in processing large data sets and generating accurate, multimodal outputs such as images or graphics, prone to hallucinations and inefficient computational resource usage due to input size constraints and lack of understanding of endpoint data structures.

Innovation Solution

An LLM interface layer determines a schema for data sets, generates prompts to guide LLMs in processing only relevant data types, and executes API calls to generate visual content efficiently, reducing hallucinations and computational load.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If LLMs process large data sets directly, then comprehensive analysis is achieved, but input size constraints and computational resource requirements make this impractical

Engineering Contradiction:
Improvedata input sizeVSAvoidprocessing efficiency
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

The patent segments the data processing workflow into distinct components: schema determination, semantic data extraction, and LLM prompt generation. This allows the system to process large data sets by working with structured schemas and extracted semantic data rather than raw data volumes, making the processing manageable within LLM input constraints while maintaining comprehensive analysis capabilities

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary processing layer that extracts semantic data from large data sets and generates structured prompts before feeding them to the LLM. This intermediary step acts as a mediator that transforms unwieldy large data inputs into concise, structured representations that the LLM can process efficiently while still capturing the essential information needed for comprehensive analysis

Inventive Principle:
Principle #24Intermediary (Mediator)

2Adaptability or versatility

If LLMs generate multimodal outputs based on large inputs, then output capability is enhanced, but hallucination increases and accuracy decreases

Engineering Contradiction:
Improveoutput capabilityVSAvoidoutput accuracy
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent performs preliminary actions by determining the schema and extracting semantic data from the data set before generating the LLM prompt. This preliminary structuring of information provides a reliable foundation that guides the LLM's multimodal generation, reducing the likelihood of hallucinations while maintaining the versatility to generate various output types based on the pre-processed semantic understanding

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system incorporates feedback mechanisms where the determined schema and extracted semantic data serve as ground truth that validates and guides the LLM's generated outputs. This feedback loop ensures that multimodal outputs remain faithful to the source data and schema constraints, improving accuracy while preserving output versatility

Inventive Principle:
Principle #23Feedback

3Measurement precision

If LLMs process more data to improve accuracy, then output fidelity increases, but computational resource consumption increases

Engineering Contradiction:
Improveoutput fidelityVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent extracts only the essential semantic data and schema information from the large data set, separating the critical information needed for accurate output from the voluminous but non-essential raw data. This extraction approach maintains output fidelity by preserving key relationships and meanings while dramatically reducing the computational burden on the LLM, as it processes only the extracted semantic representation rather than the entire data set

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP4582968A1Efficient generation of application programming interface calls using language models, data types, and enriched schema
Publication Date: 2025.07.09 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4582968A1 patent drawingFigure 1
  • EP4582968A1 patent drawingFigure 2
  • EP4582968A1 patent drawingFigure 3

AI summary

Various embodiments of the technology described herein cause an LLM to intelligently process data based on a user query and a schema determined for a data set. Certain embodiments programmatically leverage an LLM and utilize its output based on a user query. In this manner, data is processed without the LLM having to access an entire data set, and instead only utilizes information associated with the user query and the schema. The schema comprises a textual description, such as a string of alphanumeric characters, that describes the data, data types, and/or data structure of the data set. Embodiments of the technology described herein are performed by an LLM interface layer separate from a user device layer and an LLM layer. The LLM interface layer is positioned between an LLM abstraction layer and an application layer by which a user can interface with the LLM interface layer.