Retrieval-Augmented Generation With Classifier-Routed Data Domains
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Retrieval-Augmented Generation (RAG) processes struggle with complex data storage systems, including multiple data domains, formats, and access permissions, leading to inefficient and irrelevant responses.
Innovation Solution
A method to generate a training dataset for a classifier that determines the appropriate retrieval process from a plurality of processes based on input prompts, using embeddings and machine learning to select the optimal data retrieval method, considering data sources, domains, formats, and access permissions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If a RAG process provides access to all data stored by an organisation as a single data source, then the LLM can access comprehensive information, but the system cannot handle complex data organisation, access rights, and multiple data domains effectively
Solution Approach 1:
The patent segments the organisation's data into multiple distinct data domains (e.g., HR, finance, operations) with different access rights and organisational structures. Each data domain is treated as a separate entity with its own retrieval process, allowing the system to manage complexity while maintaining comprehensive data access through modular organisation.
2Loss of information
If the RAG process retrieves data from multiple data domains, then more relevant information can be found, but it is unclear which data domain should be used for a given prompt
Solution Approach 1:
The patent introduces an intermediary classification layer that receives the user prompt and determines the appropriate data domain before data retrieval occurs. This intermediary classifier acts as a mediator between the prompt and multiple data domains, routing queries to the correct domain based on the prompt's content and intent, thereby solving the ambiguity of domain selection.
3Measurement precision
If different retrieval methods are used for different data formats, then retrieval accuracy improves, but the system complexity increases due to multiple retrieval processes
Solution Approach 1:
The patent creates a universal retrieval system that can handle multiple data formats (structured, unstructured, semi-structured) through a single multi-functional architecture. The system uses a unified embedding-based search mechanism that works across different data types, eliminating the need for separate retrieval processes for each format while maintaining high retrieval accuracy.
4Reliability
If access permissions are enforced for different users, then data security improves, but the retrieval process becomes more complex
Solution Approach 1:
The patent implements local quality by applying different access permission rules to different data domains based on user roles and responsibilities. Each data domain has its own access control characteristics tailored to the specific security requirements of that domain, allowing the system to enforce comprehensive security policies while maintaining simplicity through domain-specific rather than system-wide complex access control.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
Methods and apparatus for generating a response to an input prompt are provided, in which a classifier is used to determine a retrieval process, from a plurality of retrieval processes, for use in generating a response to the input prompt. Methods and apparatus are also provided for training a classifier for determining a retrieval process, from a plurality of retrieval process and for generating a training dataset for training the classifier.