Bristows AI Document Chatbot

Appearance

Requirements ​

Functional Requirements ​

1. User Authentication and Authorization ​

  • Implement OAuth2 for secure authentication using email and password.
  • Single Sign On (SSO) will be implemented post-appointment of service principle on Azure, to bypass this blocker initially.

2. File Management ​

  • Users can upload, view, and delete documents in .pdf and .docx formats, initially with potential to include other file types like images.
  • Support for handling documents of 100-120 pages per single upload, multiple batches might be required.
    • Ability to batch requests for larger documents, though this may result in slower and more expensive requests.

3. AI-Powered Chat Interface for Document Interaction ​

  • Users can interact with the AI using a chat interface to ask questions and receive responses based on the uploaded documents.
  • Two chat modes:
    • General questions with responses based on pre-trained models.
    • Specific queries where responses utilize data from user's uploaded and vectorized documents.
  • Users can view their chat history.

4. Prompt Engineering Guidance System ​

  • Provide examples of prompts.
  • Enable users to create and reuse their custom prompt templates.

5. Data Isolation ​

  • Individual user data isolation where only the respective user can access their documents and chat data.

6. Admin Panel ​

  • Basic functionalities to add, remove, and edit user profiles.
  • Admins can manage user access (invite/revoke) and edit all user details.
  • Visibility over all user interactions on the platform.

7. Integration with Azure Document Intelligence ​

  • Utilize Azure Document Intelligence for document extraction and analysis.

8. Integration with Azure Open AI and Alternative LLMs ​

  • Initially integrate with Azure Open AI.
  • Design system to allow easy swapping with other pre-trained LLMs like Mistral.
  • Consider the cost and compliance impacts of hosting open-source models.

9. Integration with External Email Server ​

  • Support for user registration, verification, and password reset functionalities.

10. Document and Model Updates on Deletion ​

  • Ensure that when a document is deleted, the associated vectorized data and model knowledge are updated to reflect the removal.

11. Encryption of Data ​

  • All stored and transmitted data must be encrypted to ensure security and privacy.

12. Dashboard for Token and Cost Management ​

  • Provide a dashboard to monitor and display API token consumption.
  • Include features to estimate data ingestion and interaction costs prior to executing operations.