Skip to content

RAI_FAQ.md

Latest commit

 

History

History
30 lines (23 loc) · 3.3 KB

RAI_FAQ.md

File metadata and controls

30 lines (23 loc) · 3.3 KB

Azure Language OpenAI Conversational Agent Accelerator: Responsible AI FAQ

  • What is Azure-Language-OpenAI-Conversational-Agent-Accelerator?

    • The Azure-Language-OpenAI-Conversational-Agent-Accelerator project provides users with a code-first example on how to augment an existing RAG solution with Azure AI Language functionality. The system takes as input user chat messages (grounded within a specific context). The system orchestrates user chats to Azure AI Langauge CLU or CQA projects, or to a fallback RAG agent. Output is a system chat message, which may be the answer to a question in a CQA project, the result of a linked intent in a CLU project, or a general grounded response from RAG.

  • What can Azure-Language-OpenAI-Conversational-Agent-Accelerator do?

    • Overall, the demo provided with this proejct showcases the following chat experience:
      • User inputs chat dialog.
      • AOAI node preprocesses by breaking input into separate utterances.
      • Orchestrator node routes each utterance to either CLU, CQA, or fallback RAG.
      • If CLU was called, call extended business logic based on intent/entities.
      • Agent summaries response (what business action was performed, provide answer to question, provide grounded response).

  • What is Azure-Language-OpenAI-Conversational-Agent-Accelerator's intended uses?

    • The Azure-Language-OpenAI-Conversational-Agent-Accelerator project is intended to display the "better together" story when using Azure AI Language and Azure OpenAI. This chat experience includes single-turn chatting, QA, and groundedness. When combined with an existing RAG solution, adding a UnifiedConversationOrchestrator object can help in the following ways:
      • manual overrides of DSAT examples using CQA.
      • extended chat functionality based on recognized intents/entities using CLU.
      • consistent fallback to original chat functionality with RAG.

  • How was Azure-Language-OpenAI-Conversational-Agent-Accelerator evaluated? What metrics are used to measure performance?

    • Azure-Language-OpenAI-Conversational-Agent-Accelerator underwent red teaming RAI procedures to ensure metrics of harmful content and groundedness were met.

  • What are the limitations of Azure-Language-OpenAI-Conversational-Agent-Accelerator? How can users minimize the impact of Azure-Language-OpenAI-Conversational-Agent-Accelerator's limitations when using the system?

    • System is not intended for use in the context of sensitive topics or harmful content. Ensure all system content is in the context of linked project data (e.g. Contoso Outdoors).

  • What operational factors and settings allow for effective and responsible use of Azure-Language-OpenAI-Conversational-Agent-Accelerator?

    • Users provide their own AOAI resource. Depending on the AOAI model provided, accuracy of Azure-Language-OpenAI-Conversational-Agent-Accelerator may vary. If you choose to alter the provided example data, ensure that it meets guidelines of responsible AI practices.

  • How do I provide feedback on Azure-Language-OpenAI-Conversational-Agent-Accelerator?

    • Please submit feedback through the GitHub repo by creating an issue. Issues will be triaged and addressed in a timely manner. You may also contact the team at taincidents@microsoft.com.