Skip to content

PowerMind FeatureHub User Guide

This guide describes how to navigate and operate the FeatureHub Web Console graphical user interface.

Header and Workspace Selector

The header of the Web Console contains critical context controls: - Workspace Selector: Displays the current workspace name. Click the dropdown to switch between workspaces. Switching workspaces immediately reloads the UI to display only the datasources, feature views, services, exports, and keys owned by that workspace. - Workspace ID: Shown next to the workspace selector for copy-pasting into code or SDK clients.

1. Dashboard

The Dashboard is the overview page. It provides: - Summary Statistics: Total workspaces, registered data sources, feature views, and active feature services. - Recent Activity: A chronological list of recent updates, creations, or materializations across all resources, giving you an immediate view of changes in your workspace.

2. Workspaces

The Workspaces page acts as your project management hub: - New Workspace: Click this button to create a new workspace. You will enter a name and optional description. Once created, the UI automatically switches your active workspace to the new one. - Delete Workspace: Deletes the workspace.

[!WARNING] Deleting a workspace is destructive and irreversible. It removes all workspace-level uploads, exports, materialized feature data, and metadata registries.

3. Data Sources

The Data Sources page is where you connect raw input datasets.

Registering Data Sources

Click Add Data Source and select one of the following types: 1. CSV Upload: Enter a name and drag/drop or choose a .csv file from your computer. The file is uploaded and stored in workspace-managed storage. 2. PostgreSQL / MySQL: Select the database type, enter a connection URI (e.g. postgresql://user:password@host:port/database or mysql://user:password@host:port/database), and the table name. Passwords are masked upon return.

Preview and Schema Suggestions

  • View Preview: Click View on any data source to inspect the first few sample rows. Use this to verify that columns look correct and that timestamps are readable.
  • Schema Suggestions: Click Schema to see inferred column types, recommendations for join keys, and event timestamp column suggestions.

4. Feature Views

Feature Views define how raw columns from data sources map into features.

Creating a Feature View

  1. Click New Feature View.
  2. Select a registered data source.
  3. Enter a unique base name.
  4. Review the suggested Join Key and Event Timestamp Column mappings.
  5. Select the stored columns you want to import.
  6. Add derived features (derived at ingest time via SQL expression) or apply scaling/standardization transformations if necessary.
  7. Click Preview to inspect the resulting table shape.
  8. Click Create Feature View to save.

Materialization and Versioning

  • Materialize: Click Materialize on a view that has stored features. This loads the values from offline storage into the low-latency online database. Real-time inference will return empty results for stored features until the view is materialized.
  • Activation: Feature views are versioned (e.g., customer-activity__v1). Click Activate to change which version of a feature view is the default active version for predictions.

5. Feature Services

Feature Services package groups of versioned feature views and individual features into named contracts for models.

Creating a Feature Service

  1. Click New Feature Service.
  2. Enter a name (e.g., loan_eligibility_service) and optional description.
  3. Select a feature view, and check the specific feature fields your model requires.
  4. Click Create Service.

Like feature views, services are versioned (e.g., loan-eligibility-service__v1). You can toggle which version is active by clicking Activate in the services table.

6. Online Serving

The Online Serving interface provides a mock environment to test feature lookups: 1. Select a Feature Service. 2. Click Prefill Example to automatically populate the JSON fields with the correct schema structure. 3. Enter entity IDs (e.g. {"customer_id": "C001"}) in the Entity Rows panel. 4. If the service expects real-time client inputs, enter them in the Request Features panel. 5. Click Service Online Lookup to see the resulting feature values that would be returned to your model during a live prediction.

7. Exports

Exports allow you to create downloadable snapshots of your feature views for offline training: 1. Go to Exports. 2. Select a feature view. 3. Click Create Export. 4. The system processes the request in the background. Once the status changes to ready, click Download to retrieve the CSV file.

8. Integration Keys

Integration Keys are API keys used to authenticate automated scripts, pipelines, and external serving services (like model APIs) without giving them user password credentials: 1. Click Generate Key. 2. Enter a descriptive name (e.g., Inference Server Key) and an optional expiration. 3. Click Generate Key. 4. Copy the plaintext API key immediately.

[!IMPORTANT] The key is only shown once. If you close the modal, you cannot retrieve it again and must generate a new one.

To disable access, click Revoke in the integration keys table.