Skip to main content

Model Execution Workflow

The Model Execution Workflow on the Gesund.ai Platform is built to handle prediction tasks seamlessly — from individual image inference to large-scale batch predictions and the recovery of failed jobs. This ensures a reliable and efficient AI deployment experience for all users.

Model Execution Workflow

1. Initialization​

  • The platform initializes necessary components and system resources.
  • This includes model loading, environment setup, and configuration validation for prediction readiness.

2. Single Image Prediction​

When a single image prediction is requested:

  • Request Handling: The system receives the request and extracts image and dataset IDs.
  • Preprocessing: Input data is loaded and any user-defined transformations are applied.
  • Inference: The selected model processes the image to generate predictions.
  • Response: Results are returned to the user promptly.

3. Batch Prediction​

For multiple image predictions:

  • Batch Intake: The system registers and queues multiple prediction jobs.
  • Environment Scaling: Workers are dynamically provisioned to match workload demands.
  • Parallel Processing: Predictions run concurrently for optimal performance.
  • Result Aggregation: Results from all tasks are combined and returned collectively.

4. Failed Batch Recovery​

If any tasks fail during batch execution:

  • Failure Detection: The platform identifies and logs failed tasks automatically.
  • Retry Mechanism: Failed jobs are retried after addressing system or data-related issues.
  • Completion Assurance: Ensures a high success rate by recovering from transient errors.

5. Result Delivery​

  • Live Updates: Users receive real-time feedback on prediction progress.
  • Output Access: Successful results are made accessible through the interface.
  • Error Reporting: When issues occur, detailed error logs help users troubleshoot quickly.

This robust execution pipeline ensures that model inference — whether for a single case or at scale — remains efficient, resilient, and transparent for all users.