LLM Visualization Server With Safe Query-to-Code Execution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data visualization tools require technical expertise, lack flexibility and scalability, and struggle with customization, context awareness, and user experience challenges, especially when handling complex metrics and user queries.
Innovation Solution
A two-step process using Large Language Models (LLMs) for generating data visualizations, including code generation and execution, with template-based code templates and pre-defined functions, and question generation to align user intent, ensuring accuracy and scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional data visualization tools are used, then data can be visualized, but users must possess technical expertise in data preprocessing, scripting, or working with specific visualization libraries
Solution Approach 1:
The patent introduces an intermediary system comprising a natural language processing module and a code generation module that translates user-friendly natural language queries into executable visualization code. This intermediary layer shields users from technical complexity while enabling sophisticated data visualization capabilities, effectively resolving the contradiction between ease of use and the technical expertise traditionally required.
Solution Approach 2:
The system employs automated code generation and execution capabilities that allow the tool to perform data preprocessing, code generation, and visualization execution autonomously based on user queries. This self-service approach eliminates the need for users to manually write scripts or understand complex preprocessing steps, significantly improving ease of operation without sacrificing functionality.
2Adaptability or versatility
If conventional data visualization tools are used, then basic visualizations can be created, but they lack the depth, flexibility, and scalability required to analyze various metrics across long-time horizons and across a wide spectrum of organizations
Solution Approach 1:
The patent implements a universal system architecture where a single platform can handle diverse data sources, multiple visualization types, and various analytical requirements through a unified code generation and execution framework. This multi-functional design enables the system to adapt to different organizations, time horizons, and metrics without requiring separate tools, thereby achieving flexibility and scalability while managing complexity through standardization.
Solution Approach 2:
The system dynamically adapts its behavior based on user queries and data characteristics. The code generation module dynamically creates appropriate preprocessing and visualization code tailored to each specific request, and the execution environment dynamically loads and runs the generated code. This dynamic approach allows the system to scale from simple to complex visualizations as needed, providing versatility without requiring users to manage system complexity.
3Reliability
If custom data visualization solutions are developed, then specific needs can be met, but development time and resource requirements increase significantly
Solution Approach 1:
The patent implements a library of pre-defined visualization templates and code patterns that have been previously developed and validated. When users submit queries, the system selects and adapts these pre-prepared templates rather than generating code from scratch, significantly reducing development time while maintaining customization accuracy through parameter adjustment and template selection.
Solution Approach 2:
The system uses template-based code generation where proven visualization patterns are copied and adapted to meet specific user needs. Instead of developing custom solutions from scratch, the system replicates and modifies established visualization templates, ensuring reliability through proven patterns while minimizing development time through reuse of existing code structures.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Method, application server, and non-transitory computer-readable medium for generating data visualizations using Large Language Models (LLMs) are disclosed. In an aspect, a user query request is received for generating a visual representation of data. Further, the user query request is validated to determine whether the user query request is safe to accept. Based upon the validation, machine executable instructions corresponding to the user query request are generated using a Fortune Analytics Language Model (FALM). The FALM is a pretrained LLM. Furthermore, the machine executable instructions are executed. Based upon executing the machine executable instructions, a machine generated visual representation of the data is generated.