RAG Retrieval Pipeline for Structured and Unstructured Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Retrieval-Augmented Generation (RAG) frameworks struggle to efficiently and accurately process both unstructured and structured data, particularly failing to utilize structured data effectively for context-aware responses.
Innovation Solution
An RAG framework that includes an RAG pre-processing platform for vector embedding of unstructured data and summary/metadata creation for structured data, combined with an RAG retriever and reader platform to retrieve and combine context-relevant information for both data types, enhancing the LLM prompt for accurate responses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If RAG uses typical embedding process and vector search for structured data, then unstructured data retrieval is successful, but structured data understanding and usage is poor
Solution Approach 1:
The patent segments the data processing pipeline into distinct pathways: unstructured data undergoes traditional embedding and vector search, while structured data is transformed into natural language summaries and metadata that are then embedded. This segmentation allows each data type to be processed according to its characteristics, resolving the contradiction between maintaining unstructured data retrieval accuracy and improving structured data understanding.
Solution Approach 2:
The patent changes the parameters of structured data by transforming tabular data into natural language summaries and extracting metadata, then applying embedding to these transformed representations. This parameter change enables structured data to be processed using the same vector search mechanism as unstructured data, improving reliability without sacrificing the established unstructured data retrieval accuracy.
2Device complexity
If RAG stores only unstructured data in vector data store, then implementation is simple, but structured data cannot be utilized
Solution Approach 1:
The patent makes the vector data store universal by enabling it to store both traditional embedded unstructured data and embedded natural language summaries of structured data. The same embedding and retrieval infrastructure serves both data types, achieving multi-functionality without significantly increasing implementation complexity.
Solution Approach 2:
The patent introduces natural language summaries and metadata as intermediaries that bridge structured data and the vector search system. These intermediaries transform structured data into a format compatible with embedding, allowing the existing vector data store infrastructure to handle structured data without major modifications.
3Reliability
If RAG retraining LLM with new data, then model has current information, but cost and time increase
Solution Approach 1:
The patent performs preliminary action by pre-computing natural language summaries and metadata from structured data sources, and pre-applying embedding to these summaries. This preparation work is done in advance, allowing the system to quickly retrieve current information without requiring time-consuming LLM retraining when new data becomes available.
Solution Approach 2:
The patent extracts only the essential information from structured data sources into natural language summaries and metadata, separating this extracted information from the full data sources. This extraction allows the system to work with condensed, essential information that can be quickly embedded and retrieved without retraining the entire model on complete data sets.
Data Source
AI summary
A Retrieval-Augmented Generation (“RAG”) framework may include an RAG vector data store with information about vector embeddings. An RAG pre-processing platform may access unstructured data and perform vector embedding to generate a vector embedding for unstructured data to be stored in the RAG vector data store. The RAG pre-processing platform may also access structured data from the knowledge base and create a summary and metadata about the structured data to be stored in the RAG vector data store. An RAG retriever platform may receive a user prompt from a user, perform vector embedding, and retrieve context-relevant information for unstructured and structured data by searching for similar embeddings in the RAG vector data store. An RAG reader platform may combine the context-relevant information with the user prompt and an RAG prompt to create a LLM prompt. A context-aware response is then output to the user.