Background
The system consists of two separate modules:
• CKB – responsible for fetching, scraping, and cleaning raw documents.
• LLM Module – receives cleaned documents and executes the full processing pipeline: chunking, metadata enrichment, embedding, vector database indexing, and RAG-based search.
There is a need to support an alternative downstream target for processed documents. After chunking and metadata enrichment, processed document chunks should optionally be pushed to Azure Blob Storage for use by Azure-native services (e.g., Azure AI Search), instead of always continuing through the LLM Module embedding and indexing pipeline.
This task covers the implementation of a configurable routing middleware within the LLM Module, including both a deployment-time backend configuration and a runtime user-facing UI control. Users will be able to choose in the application settings where RAG search is served from: Azure AI Search or the LLM Module vector database.
Goals
• Enable the LLM Module to route processed document chunks to either Azure Blob Storage or the local embedding/RAG pipeline, based on both deployment configuration and runtime user preference.
• Frontend UI component: a settings toggle/selector allowing the user to choose between Azure AI Search and LLM Module RAG as the active search backend, without requiring redeployment.
• Set a deployment-time default so that environments without Azure credentials automatically fall back to local RAG.
• Maintain full backward compatibility – existing deployments with no Azure configuration must continue to work without any changes.
• Runtime route preference stored in the application database.
Background
The system consists of two separate modules:
• CKB – responsible for fetching, scraping, and cleaning raw documents.
• LLM Module – receives cleaned documents and executes the full processing pipeline: chunking, metadata enrichment, embedding, vector database indexing, and RAG-based search.
There is a need to support an alternative downstream target for processed documents. After chunking and metadata enrichment, processed document chunks should optionally be pushed to Azure Blob Storage for use by Azure-native services (e.g., Azure AI Search), instead of always continuing through the LLM Module embedding and indexing pipeline.
This task covers the implementation of a configurable routing middleware within the LLM Module, including both a deployment-time backend configuration and a runtime user-facing UI control. Users will be able to choose in the application settings where RAG search is served from: Azure AI Search or the LLM Module vector database.
Goals
• Enable the LLM Module to route processed document chunks to either Azure Blob Storage or the local embedding/RAG pipeline, based on both deployment configuration and runtime user preference.
• Frontend UI component: a settings toggle/selector allowing the user to choose between Azure AI Search and LLM Module RAG as the active search backend, without requiring redeployment.
• Set a deployment-time default so that environments without Azure credentials automatically fall back to local RAG.
• Maintain full backward compatibility – existing deployments with no Azure configuration must continue to work without any changes.
• Runtime route preference stored in the application database.