denim2x / web-knowledge-assistant

An engaging virtual assistant capable of answering arbitrary open-ended questions

Web Knowledge Assistant

An engaging virtual assistant capable of answering arbitrary open-ended questions

Architecture

• designed for Microsoft Azure
• Python server, running on Azure App Service
• QnA Maker instance with the qna_chitchat_professional dialog pack
• Azure Storage account for Table storage
• Cognitive Answer Search for finding relevant (and succint) answers around the Web

Workflow

• the user is greeted with an introductory message from QnA Maker
• the user may submit a message (usually a question) using the input area
• the server sends the question to QnA Maker and obtains a preliminary list of answers
• if there are no relevant answers in the list, the query is first searched in the database, according to a specialized similarity metric
• if any relevant entries are found then the best answer is shown; otherwise the query is forwarded to Answer Search
• the search results are then sorted according to the same similarity metric against the query together with the previous answer's caption (if available)
• the best answer, if available, is shown to the user
• this process can be repeated indefinitely

Setup

Requirements

• <project root>/config.yaml with the following:

azure:
  storage:
    account: ...
    key: <admin key>
  qna_maker:
    host: ...
    kb: ...
    key: ...
  cognitive:
    answer_search: <api_key>

MIT License