Feature Store

Overview

The Feature Store in timveroOS provides a centralized interface for creating and managing data transformations that are used exclusively within workflow decision processes. It enables administrators to define how raw data from various sources is transformed into features for automated decision-making.

Business Context

Financial institutions require consistent data transformations across their underwriting workflows. The Feature Store addresses this need by providing a managed repository where data transformation logic is defined, tested, and made available for use in workflow configurations.

Core Functionality

Feature Definition

A feature is a data transformation that converts raw data from integrated sources into a format usable by workflow decision logic. Features can include:

• Direct value extractions (credit scores from bureau data)

• Calculated values (debt-to-income ratios)

• Derived indicators (payment pattern analysis)

• Aggregated metrics (total outstanding debt)

Feature Management Process

1. Creation: Define transformation logic from available data sources

2. Validation: Test transformations against sample data

3. Storage: Save validated features to the repository

4. Workflow Integration: Reference features in workflow decision tasks

Configuration Interface

Accessing the Feature Store

Navigate to Settings → Feature Store

Creating a Feature

Step 1: Feature Definition

The feature creation form requires the following configuration:

Required Configuration:

• Feature Name: Unique identifier for the transformation

• Description: Business purpose of the feature

• Data Source: Selected from integrated providers

• Transformation Type: Direct value or calculated

Step 2: Transformation Logic

The transformation editor supports JavaScript and Python syntax for defining transformations:

Supported Languages:

• JavaScript

• Python

Transformation Capabilities:

• Mathematical calculations

• Conditional logic

• Data type conversions

• Value mappings

• String manipulation

• Array operations

Step 3: Validation

The validation process includes:

• Select sample data from existing records

• Execute transformation

• Review output values

• Confirm expected results

Step 4: Save to Repository

After successful validation, the feature is saved to the system repository and becomes available for workflow configuration.

Feature Categories

Data Source Features

Features that extract values directly from integrated data sources:

• Credit bureau scores

• Bank account balances

• Business registration dates

• Identity verification results

Calculated Features

Features that perform calculations on source data:

• Financial ratios (DTI, LTV)

• Time-based calculations (account age, business tenure)

• Statistical aggregations (average transaction amount)

Derived Features

Features that apply business logic to create new indicators:

• Risk categorization based on multiple factors

• Eligibility flags from complex rules

• Warning indicators from pattern analysis

Integration with Workflows

Using Features in Workflow Tasks

Features created in the Feature Store become available in workflow configurations through:

1. Load Datasource Tasks

   • Select feature from available list

   • Feature value retrieved during workflow execution

   • Result available for subsequent workflow logic

2. Expression Tasks

   • Reference feature values in calculations

   • Combine multiple features for complex logic

   • Create conditional branches based on feature values

3. Save to Profile Tasks

   • Store feature values in participant profiles

   • Make values available for workflow decisions

   • Track feature values for audit purposes

Feature Store Repository

Data Storage

All feature definitions are stored in the system repository with:

• Version control for feature modifications

• Usage tracking across workflows

• Performance metrics for optimization

• Audit trail of changes

Feature Discovery

The Feature Store interface provides:

• Category filtering

Configuration Examples

Example 1: Direct Value Feature

Name: applicant_credit_score
Data Source: Experian Credit Report
Transformation: Extract value from creditScore field
Output Type: Numeric

Example 2: Calculated Feature

Name: debt_to_income_ratio
Data Sources: Credit Report, Income Verification
Transformation: total_monthly_debt / verified_monthly_income
Output Type: Percentage

Example 3: Derived Feature

Name: payment_risk_category
Data Sources: Payment History
Transformation: IF late_payments > 3 THEN "High" ELSE "Low"
Output Type: Text

Best Practices

Feature Naming

• Use lowercase with underscores

• Include unit of measurement where applicable

• Prefix with category for organization

Documentation

• Provide clear business descriptions

• Document expected value ranges

• Note any data dependencies

• Include validation examples

Performance Considerations

• Test transformation performance

• Monitor execution times

• Optimize complex calculations

• Consider caching strategies

Common Use Cases

Credit Assessment Features

• Credit score extraction and banding

• Debt obligation calculations

• Payment history analysis

• Income verification metrics

Risk Evaluation Features

• Fraud indicator detection

• Identity verification scoring

• Address stability checks

• Business viability metrics

Eligibility Determination Features

• Product qualification flags

• Regulatory compliance checks

• Portfolio concentration limits

• Documentation completeness

Troubleshooting

Feature Creation Issues

Data source not available

• Verify data source integration is active

• Check authentication credentials

• Review data source configuration

Transformation errors

• Validate syntax in transformation logic

• Check data type compatibility

• Review sample data for edge cases

Validation failures

• Ensure test data contains required fields

• Verify calculation logic

• Check for null value handling

Workflow Integration Issues

Feature not appearing in workflow

• Confirm feature is saved and active

• Refresh workflow designer

• Check feature permissions

Feature returning null values

• Verify data source connectivity

• Check transformation error logs

• Review workflow execution sequence

Technical Considerations

Performance

• Features execute during workflow processing

• Complex transformations may impact workflow speed

• Monitor feature execution times

• Optimize frequently used features

Data Governance

• Feature modifications tracked in audit log

• Version history maintained

• Usage statistics available

• Impact analysis for changes

Next Steps

With features configured in the Feature Store:

• See Workflow Fundamentals to use features in decision processes

• Explore Metrics Engine for business calculations used in covenants

• Review Covenants Setup to see how metrics are used

---

For additional Feature Store support, consult your implementation team or system documentation.