Skip to main content

Heimdall Loop

Heimdall Loop is the adaptive online learning product on Heimdall by RejiCo — models that improve incrementally from each live prediction without full batch retraining.

Product page

heimdallapp.org/product/loop

Before you start
  • Signed-in account
  • Loop uses its own wizard — does not consume Lake gold today
  • Best when predictions feed back as training signal (recommendations, moderation, adaptive scoring)

Open Loop in the app​

Sidebar → Data Intelligence → Loop (/loop)

Loop vs ML vs Forecast​

LoopMLForecast
LearningContinuous from live dataBatch train once on goldBatch train on time series
Data sourceLoop wizard schemaLake gold or samplesLake gold or AirPassengers
RetrainingAutomatic per feedbackManual retrainManual retrain
ExplainabilityFeature contributions per predictionBatch metricsForecast intervals & metrics

Pick Loop when the model must adapt as users interact — not when you have a static spreadsheet and want one-shot batch training (use ML).

Pick ML when you have labeled historical rows and want a fixed model until you retrain.

Pick Forecast when you need future values along a time axis.

→ Full comparison: What is Heimdall?

How loops work​

  1. Define — input/output schema in Loop wizard
  2. Bootstrap — initial training examples
  3. Predict — REST API with feature contributions
  4. Learn — send outcomes back as training data
  5. Repeat — model updates incrementally

FAQ​

Does Loop use Lake gold?
Not today — Loop has its own wizard and schema. Tabular batch workflows use ML or Forecast instead.

When should I choose Loop over ML?
When predictions feed back as training signal and the model must adapt continuously — recommendations, moderation, adaptive scoring. See comparison above.

How do I send training feedback?
Use the Loop train REST endpoint documented in API integration after each predict cycle.

Getting started​

  1. Loop modeling — create your first loop
  2. Deployment — API key
  3. Monitoring — per-loop Usage section

Next steps​