Support Centre

Document Copilot

In this article, you will learn:

  • What is the Document Copilot feature,
  • Why we have decided to do this way,
  • Why use the Document Copilot feature,
  • How does this feature work,
  • Data processing,
  • Early Adopters Programme.

What is the Document Copilot feature?

The Document Copilot is an AI-supported feature, using LLM (Large Language Models) to allow the meeting participant to have a quick and easy-to-read summary of a meeting document. The summary can be of different types, and the user can also ask questions about the document using natural language, and get an answer in the same way through a chatbot interface. 

It’s like having ChatGPT read the document and help the participant to digest it more easily. But in our case, the AI model is working on our infrastructure, and no data is shared outside the Sherpany context.

Why have we decided to do it this way?

During our research, we have learned that this kind of feature has been raising a lot of curiosity in our customers. But also a lot of concern due to confidentiality, compliance and security of the company information. 

To make our feature work on-premises, was a conscious decision on our side to provide a more confident solution to our users, and aligned with our security and compliance high standards.

Why use the Document Copilot feature?

The goal of this feature is to support participants during meeting preparation so that they can more easily digest large documents that are very time-consuming to read. By having a quick summary and being able to get answers to simple questions about it quickly, the user can easily get a high-level understanding of each document and decide where to invest more time in detail reading if needed. 

It can also help participants have an easier understanding of the document information if they are not familiar with the technicality of the topic, are not experts in the matter, or are not even native language speakers of the document language.

For example:

  • Support a German speaker to more easily understand a document that is written in English, and where some terminology might not be so familiar.
  • Support an Italian speaker to have a high-level summary and ask questions on the document, using Italian, even if the document is written in French and the user doesn’t understand it.
  • Support a CIO to better understand a financial report by having the AI explain some financial terminology and metrics in a more accessible way.
  • Support a legal team member to better understand an investment proposal for a project in an industry he doesn’t have expert knowledge, and know what questions to ask and what risks to consider and look for in the project.

How does the Document Copilot feature work?

On the view mode of the meeting, when a document is open, the user can see a floating button on the bottom right to open the chatbot. Opening this bot will provide a modal where you can select one of four different types of summaries:

  • Long summary - Approximately 500 words
  • Short summary - Approximately 50 words
  • Bullet-point summary - A maximum of 10 bullets
  • Action item summary - All action items extracted

You can also type a question in natural language to get an answer from the chatbot using the context of the document.

Warning

The feature will be available in the meeting documents with PDF format (or MS Office converted to PDF) and is not available on password-protected PDFs.

Data processing

  • This AI model is used only temporarily to process the user questions and doesn’t record any client data. The client data used to provide the answers, exists only for a few seconds in this model, it is encrypted in transit and is never persisted outside the usual database.
  • This feature uses Generative Artificial Intelligence in the form of a Large Language Model. However, we are not using ChatGPT from OpenAI or any other kind of external cloud service. We are using open source models that run inside our own infrastructure, and therefore, we are not exposing our client data to external entities for data processing.
  • We are not using the documents to train our AI Model: the model has no memory to remember what it processed and answered about any document and data is not leaked between different users, not even between users from the same organisation.

Early Adopters Programme

For now, the feature will be available for customers in the Early Adopters Programme.

If you want to know more about this programme, you can check this page or contact your Sherpany Key Account Manager.

Did you find it helpful? Yes No

Send feedback
Sorry we couldn't be helpful. Help us improve this article with your feedback.