ML Pro Nodes Reference
XGENIA provides five specialized pro nodes for machine learning workflows. These are visual nodes you can place on the canvas and wire into your application logic.
Available Nodes​
| Node | Purpose |
|---|---|
| Auto ML Analyzer | Analyze datasets for ML readiness |
| Auto ML Trainer | Train models with HuggingFace AutoTrain |
| Auto ML Predictor | Run inference with HuggingFace models |
| Client Retention Analyzer | Analyze customer churn and retention |
| Retention Action Engine | Generate action plans from retention analysis |
Common Inputs​
All ML pro nodes share these common inputs:
| Port | Type | Description |
|---|---|---|
hfToken | String | Your HuggingFace API key |
mlServerUrl | String | ML Coordinator server URL (default: http://localhost:3001) |
Auto ML Analyzer​
Analyzes your data before training. Detects the problem type, suggests the best target and features, identifies data quality issues, and recommends the appropriate HuggingFace task type.
Key outputs: problemType, targetColumn, suggestedFeatures, hfTaskType, dataQualityIssues
Auto ML Trainer​
Creates a HuggingFace AutoTrain project, uploads data, and starts training. Supports task types: text-classification, tabular-classification, tabular-regression, image-classification, llm-finetuning, and dreambooth.
Key outputs: hfProjectId, hfModelRepo, trainingMetrics, status
Auto ML Predictor​
Runs inference using any HuggingFace model. Supports both serverless (free) and dedicated endpoint modes. Handles text, tabular, and image model responses.
Key outputs: prediction, confidence, rawResponse
Client Retention Analyzer​
Specialized node for customer retention analysis. Sends customer data to the ML Coordinator, which identifies churn patterns, calculates churn rate, and finds risk factors ranked by significance.
Inputs​
| Port | Type | Description |
|---|---|---|
data | Array | Customer data records |
context | String | Additional context for analysis |
Analyze | Signal | Trigger analysis |
Outputs​
| Port | Type | Description |
|---|---|---|
insights | Object | Full retention analysis |
churnRate | Number | Percentage of churned customers |
riskFactors | Array | Risk factors ranked by significance |
recommendations | Array | Suggestions to reduce churn |
error | String | Error message |
Retention Action Engine​
Takes the output from the Client Retention Analyzer and generates specific, actionable retention actions. Maps risk factors to intervention strategies based on semantic analysis of each factor.
Inputs​
| Port | Type | Description |
|---|---|---|
insights | Object | Output from Client Retention Analyzer |
riskFactors | Array | Structured risk factor objects |
Generate | Signal | Trigger action generation |
Outputs​
| Port | Type | Description |
|---|---|---|
actions | Array | Generated retention actions |
actionPlan | Object | Structured action plan with priorities |
error | String | Error message |
Prerequisites​
All ML pro nodes require:
- HuggingFace API key — configured in Settings or passed via the
hfTokenport - ML Coordinator server — running at the URL specified in
mlServerUrl
See Setting Up HuggingFace for configuration details.