Interface IRAGMetric<T>

Namespace

AiDotNet.Interfaces

Assembly

AiDotNet.dll

Defines the contract for RAG evaluation metrics.

public interface IRAGMetric<T>

Type Parameters

T

The numeric data type used for relevance scoring.

Remarks

A RAG metric evaluates the quality of retrieval-augmented generation systems by comparing generated answers against ground truth or analyzing specific aspects of the generation process. Metrics help developers understand system performance and guide improvements.

For Beginners: Metrics are like test scores for your RAG system.

Think of it like grading an exam:

• The metric looks at the AI's answer
• Compares it to what the answer should be (or checks quality)
• Gives a score (0-1, where 1 is perfect)

Different metrics measure different things:

• Faithfulness: Does the answer stick to the source documents?
• Similarity: How close is the answer to the ground truth?
• Coverage: Does the answer address all parts of the question?

Use metrics to:

• Compare different RAG configurations
• Track improvements over time
• Identify weak points in your system

Properties

Description

Gets the description of what this metric measures.

string Description { get; }

Property Value

string

Name

Gets the name of this metric.

string Name { get; }

Property Value

string

Methods

Evaluate(GroundedAnswer<T>, string?)

Evaluates a grounded answer and returns a score.

T Evaluate(GroundedAnswer<T> answer, string? groundTruth = null)

Parameters

answer GroundedAnswer<T>

The grounded answer to evaluate.

groundTruth string

The expected/correct answer (null for reference-free metrics).

Returns

T

A score between 0 and 1, where 1 is perfect.