AI Integration Demo

The Velo Plot provides a powerful bridge for integrating your own AI models. This allows for real-time predictions, forecasting, and advanced signal analysis directly in the browser.

Interactive Forecasting

Below is a demonstration of an AI-powered Forecasting Tool. It uses the PluginMLIntegration to run a mock LSTM inference on historical data and visualizes both the projected trend and the uncertainty (confidence interval).

Native Regression

In-browser linear-regression training with a residual-based confidence band.

How it Works

1. Data Extraction: The plugin extracts raw X/Y data from any registered chart series.
2. External Inference: The data is passed to your registered model (which could use @tensorflow/tfjs, onnxruntime-web, or a remote API).
3. Synchronous Visualization: The resulting prediction is returned to the plugin, which renders it using the high-performance overlay system.

Example: Predictive Maintenance

Imagine monitoring laboratory equipment. You can run an anomaly detection model every few seconds:

const result = await chart.ml.runInference('anomaly-model', 'sensor-series');

if (result.metadata.anomalyScore > 0.8) {
  chart.ml.visualizeResults(result, {
    lineStyle: { color: '#ef4444' } // High alert red
  });
}

Features

• Confidence Intervals: Built-in support for visualizing prediction uncertainty.
• Low Overload: Predictions are rendered on an overlay layer, keeping the main WebGL engine focused on high-speed data updates.
• Model Agnostic: Works with any JavaScript-based ML library.