Database Modification with LLM Embeddings for Structured Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a need for methods to update and improve data in databases for various software applications, particularly in converting unstructured data into structured data and generating recommendations based on this conversion using a large language model (LLM).
Innovation Solution
The method involves receiving computing prompts, generating vector embeddings, accessing a database to identify corresponding files, receiving data and generating recommendations using an LLM, and transmitting these recommendations to a database system, thereby converting unstructured data into structured data and providing insights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If unstructured data is converted into structured data using an LLM, then data quality and database accuracy are improved, but processing time and computational resources increase
Solution Approach 1:
The patent applies preliminary action by generating vector embeddings for computing prompts before actually querying the database. This preprocessing step organizes the unstructured data into a structured format (vector embeddings) that can be efficiently searched and processed later, thereby improving data quality while managing processing time through advance preparation
Solution Approach 2:
The patent segments the data processing into distinct components: generating vector embeddings from unstructured data, storing them in a vector database, querying with new prompts, and generating recommendations. This segmentation allows each step to be optimized independently, improving overall data quality while controlling processing time through modular operations
2Measurement precision
If vector embeddings are generated and stored in a database, then data retrieval accuracy is improved, but database complexity and storage requirements increase
Solution Approach 1:
The patent introduces vector embeddings as an intermediary representation between unstructured data and the database system. Instead of storing raw unstructured data directly, the system converts it into vector embeddings that serve as a bridge, enabling accurate similarity-based retrieval while maintaining a manageable database structure
Solution Approach 2:
The patent transforms data from one parameter state (unstructured text) to another (vector embeddings with numerical parameters). This parameter change enables the database to work with structured numerical data that can be efficiently queried using similarity metrics, improving retrieval accuracy while maintaining database simplicity through standardized data formats
3Reliability
If multiple computing prompts are processed through an LLM, then recommendation accuracy is improved, but computational cost and energy consumption increase
Solution Approach 1:
The patent merges multiple computing prompts into a unified vector embedding representation stored in a single database. Instead of processing each prompt independently through the LLM, the system combines them into a structured format that can be queried together, improving recommendation accuracy through comprehensive analysis while reducing computational energy by avoiding redundant LLM processing
Solution Approach 2:
The patent creates vector embedding copies of the computing prompts that can be stored and reused in the database. Instead of repeatedly processing the same unstructured prompts through the computationally expensive LLM, the system uses the generated vector embeddings as reusable copies, maintaining recommendation accuracy while significantly reducing computational energy consumption
Data Source
AI summary
Methods and systems for producing and transmitting recommendations associated with a database using a large language model (LLM). An exemplary method includes receiving a set of computing prompts from a source. Receiving vector embeddings for the set of computing prompts from an LLM and determining using the vector embeddings, files from a database that correspond with a first computing prompt from the set of computing prompts or a second computing prompt from the set of computing prompts. The system receives first data from the LLM based on the first and second computing prompts, and the files from the database. The method generates second data comprising the first data and computing indicators. Recommendations associated with the database are generated based on the second data and transmitted to a system maintaining the database.