RAG Database Ingestion Via Replicated Message Queues
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for fine-tuning large language models (LLMs) using internal company data are resource-intensive and struggle to incorporate dynamic information from collaborative systems like drives, mail, and messenger, limiting their effectiveness in providing relevant responses.
Innovation Solution
A method for building a retrieval-augmented generation (RAG) database that collects data from multiple collaborative systems, replicates message queues, and indexes data to create a searchable database, minimizing system changes and integrating dynamic data changes, while maintaining data security and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If data is collected directly from custom message queues of collaborative systems, then data collection is simple, but system complexity increases and existing systems are disrupted
Solution Approach 1:
The patent introduces a message queue replication mechanism as an intermediary layer between collaborative systems and the data collection system. Instead of directly accessing custom message queues of collaborative systems (which would increase system complexity and disrupt existing systems), the system replicates message queues to create isolated copies for data collection. This intermediary approach maintains simplicity in data collection while avoiding direct interference with collaborative system architecture.
Solution Approach 2:
The patent implements message queue replication by creating copies of the original message queues. These replicated queues contain the same data as the source queues but can be accessed and processed independently without affecting the original collaborative systems. This copying mechanism allows straightforward data collection from the replicated queues while the original systems remain unchanged, thus reducing both system complexity and disruption.
2Device complexity
If static information is embedded and indexed to build a search database, then the database structure is simple, but it cannot incorporate dynamic information from collaborative systems
Solution Approach 1:
The patent implements a dynamic database structure that can adapt to changing data from collaborative systems. The system continuously replicates message queues and updates the search database with new data, transforming a static database into a dynamic one. This allows the database to incorporate dynamic information from collaborative systems while maintaining a relatively simple indexing structure based on message queue data.
Solution Approach 2:
The patent performs preliminary actions by pre-replicating message queues and pre-indexing data structures before dynamic information arrives. The message queue replication mechanism is set up in advance, and the database indexing structure is prepared beforehand. When dynamic information from collaborative systems becomes available, the pre-configured system can quickly ingest and index this data without requiring complex real-time structural changes, thus maintaining simplicity while achieving adaptability.
3Device complexity
If message queues are replicated to reduce system complexity, then data collection becomes easier, but data synchronization overhead increases
Solution Approach 1:
The patent implements periodic message queue replication, where data is copied from source queues to replicated queues at scheduled intervals rather than continuously. This periodic replication reduces the synchronization overhead compared to real-time continuous replication, while still maintaining relatively up-to-date data in the replicated queues. The periodic action allows the system to balance between data freshness and synchronization time investment.
Solution Approach 2:
The patent applies partial replication by selectively replicating only certain message queues or certain types of messages from the source systems. Instead of replicating all data continuously, the system replicates a subset of critical or frequently accessed data. This partial action approach reduces the overall synchronization overhead and time requirements while still providing sufficient data for effective search and retrieval operations.
Data Source
AI summary
A method retrieval-augmented generation (RAG) interacting with generative AI method is provided. The method includes collecting data from a plurality of collaborative systems, and building a database to perform vector searching by embedding and indexing the data. The collecting of the data includes replicating a custom message queue to generate a replicated message queue based on a determination that a first collaborative system among the plurality of collaborative systems has the custom message queue, and collecting data from the replicated message queue instead of the custom message queue.