Get in Touch
 Duration 7 hours

Course Outline

  • Day 1: From document collection to knowledge agent
    • How AI answers from documents
      • Searching the collection and providing answers with cited sources
      • Understanding why answers may be missing or incorrect
      • Distinguishing between uploaded files and connected file storage
      • Exercise: Identify three typical questions from your domain that the agent should be able to answer.
    • Preparing knowledge
      • Managing structure, naming conventions, and content currency
      • Eliminating duplicates and conflicting versions
      • Verifying access permissions
      • Exercise: Review your document collection and create a cleaned selection.
    • Choosing a platform
      • Implementing a Copilot agent on SharePoint
      • Using a Claude project or custom GPT
      • Utilizing Langdock assistant and NotebookLM
      • Evaluating criteria such as data storage, sharing capabilities, collection size, and cost
      • Exercise: Select the most suitable platform for your specific case and provide a justification.
    • Building the agent
      • Defining instructions regarding audience, tone, and answer format
      • Ensuring every answer includes a source reference
      • Managing questions where no answer exists in the collection
      • Exercise: Construct the knowledge agent on the selected platform.
    • Testing with a question catalog
      • Developing 20 test questions with expected answers
      • Rating responses as correct, partially correct, incorrect, or missing
      • Making targeted improvements to instructions and the document collection
      • Exercise: Test the agent with 20 questions and document the hit rate.
    • Maintenance and operation
      • Assigning responsibilities for content management
      • Managing updates and version status
      • Gathering feedback from users
      • Exercise: Draft a maintenance plan outlining responsibilities and review cadence.
    • Outcome: Each participant will have a shared knowledge agent based on their documents, a measured hit rate, and a maintenance plan.

Requirements

  • Access to Copilot, Claude, ChatGPT, Langdock, or NotebookLM, along with a curated collection of finalized documents from your domain.
  • No programming experience is required.

Target Audience

  • Teams, knowledge owners, and key users seeking to provide answers sourced directly from their internal documents, policies, and manuals.

Number of participants


Price per participant

Upcoming Courses

Related Categories