Motivation

The Upload & Ask interface allows users to upload documents that are not part of the existing knowledge base or to limit the search scope to specific documents. This functionality is crucial when you need to focus exclusively on certain documents for precise and relevant information retrieval. The primary motivation is to enable users to chat exclusively with the uploaded document(s), ensuring that the responses are based solely on the content of these documents. This feature is particularly useful for summarizing large documents that exceed the context window of large language models, as it allows for multiple rounds of summarization to handle extensive content effectively.

Goal

The goal of the Upload & Ask assistant is to facilitate in-depth conversations and queries against specific documents uploaded by the user. By restricting the AI's responses to the content of these documents, users can obtain precise and relevant information, enhancing the accuracy and focus of their interactions.

Structure and Logic of Assistant

1. Document Upload:

   • Users can upload one or multiple documents directly through the chat interface.

   • These documents are then ingested by the system.

     

2. Ingestion and Processing:

   • The uploaded documents are chunked appropriately for processing.

   • The content is semantically embedded to facilitate effective searching and summarization.

3. Functionality:

   • Document Search: Users can perform a semantic and full text search within the uploaded documents to find specific information, similar to Internal Knowledge Search with the limitation of the scope.

   • Document Summarisation: Users can request summaries of the uploaded documents, with the system capable of handling long documents through iterative summarisation processes.

Possible Adaption of Assistant

The Internal Knowledge Search assistant can be adapted and customised:

• Enhanced Prompting: The system can be tailored by enhancing the prompts.

Required and optional modules

The following modules are required/optional for this assistant:

Ingestion Configuration for Assistant

Space admins can define an ingestion configuration for the Upload & Ask assistant on the space management page in the Advanced Settings section (note that by default the ingestionConfig is empty, e.g. {} like in the following screenshot):

When a user uploads a file into a space where an ingestionConfig is defined, the backend receives and applies the ingestionConfig.

Example AI Assistant Configuration:

Download the provided TXT file and upload it into a new space as an AI Assistant configuration.

• Note: By setting "scopeIds": [ "UPLOAD_ONLY" ], we limit the response to only the uploaded documents. If the answer should include documents of a specific scope in addition the the uploaded once, adjust this parameter accordingly. You will need to set this parameter for Document Summarizer and Document Search modules.