AI Query Generation for Proprietary Data Retrieval Syntax
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Legacy computer systems, particularly those with proprietary data formats and protocols, pose challenges for interoperability with modern systems due to the high effort required for implementing general-purpose interfaces, and existing language models are limited in handling complex databases with vast amounts of data.
Innovation Solution
A system utilizing a trained or fine-tuned language model to generate queries for proprietary data retrieval systems, leveraging an information schema and proprietary query syntax, enables seamless data access with reduced implementation effort and maintains security by not directly accessing the data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a general-purpose interface is implemented to support the entire functionality of the legacy computer system, then interoperability with modern systems is improved, but the implementation effort becomes punitively high
Solution Approach 1:
The patent segments the interface functionality into multiple specialized language models, each trained on specific proprietary query syntaxes and data formats. Instead of one comprehensive interface, multiple focused models handle different legacy systems separately, reducing the complexity of implementing and maintaining a single general-purpose interface while preserving interoperability capabilities.
Solution Approach 2:
The patent introduces language models as intermediary components between modern systems and legacy computer systems. These language models translate natural language queries into proprietary query syntaxes, acting as mediators that eliminate the need for direct complex interface implementations between modern and legacy systems.
2Productivity
If existing language models are used to query databases, then basic data retrieval is possible, but they are limited in handling complex databases with vast amounts of data
Solution Approach 1:
The patent applies preliminary action by training language models in advance on specific proprietary query syntaxes, data formats, and schema structures of legacy systems. This pre-training equips the models with the necessary knowledge to handle complex queries on vast databases without requiring real-time adaptation, thereby improving both productivity and adaptability.
Solution Approach 2:
The patent changes the parameters of language models by fine-tuning them on domain-specific data including proprietary query syntaxes, information schemas, and legacy system documentation. This parameter adjustment transforms general-purpose language models into specialized models capable of handling complex database structures and vast amounts of data effectively.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Various examples of the present disclosure relate to a system, method, and computer program for retrieving relevant information from a proprietary data retrieval system (401), and to a method and computer program for training or fine-tuning a language model being used for retrieving the relevant information from the proprietary data retrieval system (401). A system (40) for retrieving relevant information from a proprietary data retrieval system (401) comprises one or more processors (44) and one or more interfaces (42), wherein the system (40) is configured to obtain, via an interface, a request of a user for providing information, wherein the request is worded in natural language, provide a prompt as input for a language model, the prompt comprising information on the request, wherein the language model is trained or fine-tuned to output at least one of a query for the proprietary data retrieval system (401) according to a proprietary query syntax of the proprietary data retrieval system (401) and textual information about one or more possible follow-up requests, wherein the language model is trained or fine-tuned based on an information schema of the proprietary data retrieval system (401) and/or wherein the system (40) is configured to obtain information on the information schema based on the request and provide the information on the information schema as input for the language model, if the output of the language model comprises a query, validate the query based on the proprietary query syntax of the proprietary data retrieval system (401) and retrieve information from the proprietary data retrieval system (401) using the query, and provide a response for the user having requested the information, wherein the response comprises, depending on the output of the language model, at least one of a representation of the information retrieved from the proprietary data retrieval system (401) and a representation of the one or more possible follow-up requests.