SmartSpace.ai Workflows & Use Case Setup

Guide to setting up use cases in SmartSpace.ai: Data space creation, workspace configuration, access control, and testing for precise contenSmartSpace.ai Workflows & Use Case Setupt generation.

Use Case: Marketing Workspace

Setup Guide: Marketing Workspace in SmartSpace.ai

Step 1: Create Data Spaces

  1. General Information Data Space: Set up a data space for general marketing materials, such as your styling guide and marketing strategy. This serves as a foundational repository.

Step 2: Workspace Configuration

  1. Link Data Spaces: Link your newly created data space to the workspace.
  2. Base Settings Configuration:
    • Top P: Set to 0.2 for staying on-topic.
    • Temperature: Start with 1.5 for a balance of uniqueness and relevance.
    • Presence Penalty: Set to 0 to avoid repetitive content.
    • Frequency Penalty: Keep at 0 to prevent the creation of unfounded information.
  3. Pre-Prompt Base Model Crafting: Develop a pre-prompt that guides the LLM to produce structured, verifiable, and authentic content. It should include:
    • A clear objective.
    • Instructions to utilize only provided information.
    • A directive against creating new information.
    • A defined structure for the desired content.
    • Guidance for handling informational gaps.
    • A focus on verifiability and authenticity.

Step 3: Set Up the Collector

  1. Base Settings for Collector:
    • Top P, Presence Penalty, Frequency Penalty: Usually, default settings are appropriate.
    • Temperature: Set to 0 for consistency in document retrieval.
  2. Collector Pre-Prompt: Craft a pre-prompt that directs the Collector to retrieve specific documents relevant to your marketing goals.

Step 4: Access Control and Workspace Utilization

  1. Link User: Manage user permissions and access to the workspace.
  2. Test and Adjust: Query the workspace, refine pre-prompts, and adjust temperature incrementally for varied but relevant responses. Utilize the Sandbox environment for testing.

Example Pre-Prompt for a Press Release – Base Model

  • Develop a structured pre-prompt focusing on documented sources, prohibition of fabricated details, and a specific content structure.
  • Example:
  • Objective: Draft a press release for Hayward Construction, closely following the provided marketing strategy and styling guidelines.

    Adherence to Documented Sources: Use only information from official Hayward Construction documents or provided by the user. Ensure authenticity and relevance.

    Avoid Fabrication: Do not create quotes or attribute statements to unnamed individuals. Maintain verifiability and authenticity at all times.

    Handling Informational Gaps: For required elements not in the documents, like quotes, use generic but aligned statements reflecting the company’s values. Avoid inventing content.

    Content Structure:

    • Company Background: Give a concise overview of Hayward Construction, based on official sources.
    • Project Overview: Describe ongoing and future projects, goals, and partnerships as per the provided materials.
    • Strategic Focus: Emphasize strategic initiatives, linking them to industry trends, as documented.
    • Call to Action: Conclude with a CTA that is supported by the content and invites engagement with Hayward Construction.

    Execution Guideline: The press release should be informative, mirror Hayward Construction’s strategic messaging, and engage the audience. It must strictly adhere to the marketing and styling guidelines, avoiding fictional content, especially fabricated quotes, to ensure credibility.

Marketing Workspace Findings and Adjustments

  1. Importance of Pre-Prompt: Ensuring a well-crafted pre-prompt is crucial for consistent and accurate results. It should define the structure of the desired content.
  2. Top P and Temperature Adjustments: These should be carefully balanced. A low Top P with a higher Temperature (around 1.5) tends to work well for structured content.
  3. Presence Penalty and Frequency Penalty: Generally, these should be set to 0 to avoid unintended repetition and creation of new details.
  4. Response Variability and HTML Styling: Encourage variability and, if necessary, use HTML and CSS styling directives to enhance presentation.
  5. Potential Caching Issue: Be aware of potential caching when querying the LLM in quick succession. Wait a bit between queries for varied results.
  6. Caution with Settings: Be mindful not to set Top P at 1 and Temperature at high levels simultaneously, as this combination can result in overly creative outputs and potentially disrupt the normal functioning of the system.

Additional Recommendations

  • Use Case Specificity: Different content types like press releases, LinkedIn posts, etc., might require specific workspaces or adjustments.
  • Company Information: Having comprehensive company information and examples can greatly aid in creating effective pre-prompts.
  • Continuous Testing and Improvement: Regularly test the workspace and adjust configurations to maintain quality and relevance of the output.
keyboard_arrow_up