Natural-Language Product Configuration With Constraint Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing product configurators require domain-specific graphical user interfaces and specialized technical knowledge, limiting their usability for non-experts and incurring high development costs due to custom GUIs, and pre-trained large language models lack sufficient domain knowledge and reliability in generating feasible configurations.
Innovation Solution
A method integrating large language models with a constraint engine to translate natural-language inputs into constraints, using few-shot prompting and an intermediary JSON format, ensuring technical feasibility through a configuration engine, and providing context-aware explanations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a custom graphical user interface is designed for each product and user group, then the usability for targeted users is improved, but the development effort and cost increase significantly
Solution Approach 1:
The patent applies a universal natural language interface that works across all product lines and user groups without requiring custom GUIs for each combination. The system uses a single product-line-agnostic natural language interface that can be configured for different products through parameter definitions, eliminating the need to design separate interfaces for each product-line/user-group combination while maintaining usability through domain-specific parameter handling
Solution Approach 2:
The patent introduces a natural language interface as an intermediary layer between users and the complex configuration system. This intermediary translates user-friendly natural language queries into formal configuration constraints, shielding users from the underlying complexity of product parameters and constraints while maintaining accurate configuration capabilities
2Ease of operation
If pre-trained large language models are used for product configuration, then the ease of operation for non-experts is improved, but the reliability of configuration feasibility deteriorates due to hallucination
Solution Approach 1:
The patent implements a feedback mechanism where the constraint solver validates and corrects the LLM's generated configurations. The system receives the LLM's natural language response, extracts the proposed configuration, checks it against product constraints using the constraint solver, and provides feedback to correct any feasibility issues, ensuring reliable configurations while maintaining ease of use
Solution Approach 2:
The patent introduces a constraint solver as an intermediary verification layer between the LLM and the final configuration output. This intermediary component checks the feasibility of configurations generated by the LLM and ensures they satisfy all product constraints, preventing hallucination-related errors while preserving the user-friendly natural language interface
3Manufacturing precision
If domain-specific graphical user interfaces are used, then the manufacturing precision of product configuration is improved, but the ease of manufacture deteriorates due to custom implementation requirements
Solution Approach 1:
The patent creates a universal configuration system that handles multiple product lines through a single unified architecture. The system uses a common natural language interface and constraint solving framework that can be applied across different products by simply updating product-specific parameter definitions and constraints, eliminating the need to manufacture custom GUIs for each product line while maintaining configuration accuracy
Data Source
Figure 1
AI summary
The present disclosure provides a method for generating a product configuration. The method includes receiving a natural-language description of product requirements from a user (102), translating the natural-language description (104) to constraints using a formalizer (108) that includes a large language model, determining a product configuration (120) that satisfies the constraints and predefined product line constraints using a configuration engine (118) that includes a constraint solver, and generating a natural-language description of the product configuration (126) using an interpreter (124) that includes a large language model. The disclosure further provides a system for generating a product configuration that includes the formalizer (108), the configuration engine (118), and the interpreter (124). Additionally provided are a device for planning and manufacturing a product based on the product configuration and a computer program product for performing the method.