Custom Fields

Add extra fields to model_data_logs for custom tracking and monitoring needs.

Overview

The custom_fields parameter allows you to include additional fields in the encrypted model_data_logs sent to the RAIT API. These fields are merged with standard fields like prompt_id, ethical_dimensions, and telemetry data.

Single Evaluation

Use custom_fields to pass a dictionary of additional values:

from rait_connector import RAITClient

client = RAITClient()

result = client.evaluate(
    prompt_id="123",
    prompt_url="https://example.com/123",
    timestamp="2025-12-11T10:00:00Z",
    model_name="gpt-4",
    model_version="1.0",
    query="What is AI?",
    response="AI is artificial intelligence...",
    environment="production",
    purpose="monitoring",
    custom_fields={
        "user_id": "user_123",
        "session_id": "session_abc",
        "custom_metric": 0.95,
        "api_version": "v2.1",
        "region": "us-east-1"
    }
)

Batch Evaluation

For batch evaluation, custom_fields is specified per prompt:

from rait_connector import RAITClient

client = RAITClient()

prompts = [
    {
        "prompt_id": "batch-001",
        "prompt_url": "https://example.com/prompt/001",
        "timestamp": "2025-12-11T10:00:00Z",
        "model_name": "gpt-4",
        "model_version": "1.0",
        "query": "What is AI?",
        "response": "AI is...",
        "environment": "production",
        "purpose": "monitoring",
        "custom_fields": {
            "user_id": "user_123",
            "session_id": "session_abc",
        },
    },
    {
        "prompt_id": "batch-002",
        "prompt_url": "https://example.com/prompt/002",
        "timestamp": "2025-12-11T10:01:00Z",
        "model_name": "gpt-4",
        "model_version": "1.0",
        "query": "What is ML?",
        "response": "ML is...",
        "environment": "production",
        "purpose": "monitoring",
        "custom_fields": {
            "user_id": "user_456",
            "session_id": "session_def",
        },
    },
]

summary = client.evaluate_batch(prompts=prompts)

Use Cases

User Context Tracking

custom_fields = {
    "user_id": current_user.id,
    "organization_id": current_user.org_id,
    "subscription_tier": current_user.subscription
}

Performance Monitoring

custom_fields = {
    "response_time_ms": response_timer.elapsed(),
    "tokens_used": token_counter.count,
    "cache_hit": cache.was_hit()
}

A/B Testing

custom_fields = {
    "experiment_id": "exp_123",
    "variant": "variant_b",
    "feature_flags": ["new_ui", "enhanced_search"]
}

Custom Business Metrics

custom_fields = {
    "customer_segment": "enterprise",
    "request_priority": "high",
    "sla_tier": "gold",
    "cost_center": "engineering"
}

Data Flow

1. Base model_data_logs is constructed with standard fields
2. custom_fields dict is merged (if provided)
3. Complete model_data_logs is encrypted
4. Encrypted data is sent to RAIT API

Best Practices

1. Keep it simple: Only add fields you actually need
2. Avoid PII: Don't include sensitive personal information
3. Use meaningful names: Make field names descriptive and consistent
4. Document custom fields: Keep track of what each field means