> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getnetra.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Overview

> Understand Netra's evaluator framework — LLM-as-Judge, code-based, and rule-based evaluators for text, voice, and image AI systems.

Evaluators are the scoring logic that determines whether your AI system meets quality standards. They transform subjective assessments into measurable metrics — from semantic correctness and tool execution accuracy to audio naturalness and visual composition. Use them with [evaluations](/Evaluations/text-evaluations/Datasets) to build automated quality pipelines.

<Card title="Quick Start: Evaluation" icon="rocket" href="/quick-start/QuickStart_Evals">
  New to evaluations? Get your first evaluation running in minutes.
</Card>

## Evaluator Categories

Netra organizes evaluators by the modality they assess:

<CardGroup cols={3}>
  <Card title="Text Evaluators" icon="font" href="/Evaluators/text-evaluators">
    LLM-as-Judge, code-based, and rule-based evaluators for text outputs. Assess answer correctness, relevance, safety, tool execution, and custom business logic.
  </Card>

  <Card title="Voice Evaluators" icon="microphone" href="/Evaluators/voice-evaluators">
    Audio-aware evaluators for TTS, STT, and conversational delivery. Score voice naturalness, transcription accuracy, pronunciation, and speaking behavior.
  </Card>

  <Card title="Image Evaluators" icon="image" href="/Evaluators/image-evaluators">
    Multimodal and rule-based evaluators for image generation and editing. Assess composition, style, format compliance, and image-text alignment.
  </Card>
</CardGroup>

## Evaluator Types

Across all categories, Netra provides several evaluation approaches:

| Type                         | Best For                                                          | How It Works                                             |
| ---------------------------- | ----------------------------------------------------------------- | -------------------------------------------------------- |
| **LLM as Judge**             | Subjective quality, semantic correctness, nuanced criteria        | Uses AI models to evaluate AI outputs                    |
| **Audio LLM as Judge(LALM)** | Voice naturalness, pronunciation, conversational delivery         | Listens to call recordings to score audio quality        |
| **Code Evaluator**           | Deterministic checks, JSON validation, regex, custom logic        | Runs JavaScript or Python handlers                       |
| **Rule-Based Checks**        | Latency, cost, token usage, tool-call matching, format validation | Configured through forms without writing prompts or code |
| **Multimodal Judge**         | Image quality, composition, style, text-image alignment           | Uses vision-capable models to assess visual outputs      |

## Cross-Modal Evaluation

Evaluators are not limited to their primary modality. Text evaluators can evaluate transcript content from voice interactions, making them useful for scoring conversation quality even when the input is audio-derived.

| Scenario                           | How to Evaluate                                                                                                     |
| ---------------------------------- | ------------------------------------------------------------------------------------------------------------------- |
| Voice agent conversation quality   | Use voice evaluators for audio quality + text evaluators on the transcript for factual accuracy and topic adherence |
| Voice agent goal completion        | Run text-based goal fulfillment evaluators against the conversation transcript                                      |
| Voice agent safety                 | Apply text-based guardrail evaluators (toxicity, bias) to the transcript                                            |
| Image generation with text prompts | Use text evaluators to assess prompt adherence alongside image evaluators for visual quality                        |

<Info>
  When a voice simulation completes, Netra produces both audio and transcript data. You can attach text evaluators to the same evaluation to evaluate the conversation content, while voice evaluators assess the audio delivery — giving you full coverage of both what was said and how it sounded.
</Info>

## Evaluator Library

Netra provides a library of 49 preconfigured evaluators across 12 categories, ready to use or customize:

| Category            | Modality | Representative Evaluators                                             |
| ------------------- | -------- | --------------------------------------------------------------------- |
| **Quality**         | Text     | Answer Correctness, Answer Relevance, Faithfulness, Context Precision |
| **Agentic**         | Text     | Goal Accuracy, Goal Fulfillment, Information Elicitation              |
| **Guardrails**      | Text     | Toxicity, Bias, Topic Adherence                                       |
| **Performance**     | Text     | Latency, Cost, Token Usage                                            |
| **Semantic**        | Text     | Semantic Similarity, SQL Semantic Equivalence                         |
| **Tool Use**        | Text     | Tool Correctness (exact, partial, sequence)                           |
| **JSON Evaluator**  | Text     | JSON key/value validation                                             |
| **Regex Evaluator** | Text     | Pattern matching                                                      |
| **Multimodal**      | Image    | Image Analysis, Image Editing Quality, Image-Text Coherence           |
| **TTS**             | Voice    | Humanness, Pronunciation Accuracy, Speaking Rate                      |
| **STT**             | Voice    | Word Error Rate, Transcription Correctness, Time to First Transcript  |
| **Conversational**  | Voice    | Backchannelling Naturalness, Expressiveness, Speaking Percentage      |

## Creating Evaluators

You can create evaluators in two ways:

1. **From the Library** — Browse prebuilt evaluators, click **Add**, customize the prompt or parameters, and save to **My Evaluators**
2. **Custom** — Click **Add Custom**, choose the evaluator type (LLM as Judge, Code, or rule-based), and configure from scratch

Every evaluator supports Playground testing so you can validate it against sample data before attaching it to evaluations.

## Getting Started

<Steps>
  <Step title="Choose Your Modality">
    Decide whether you're evaluating text, voice, or image outputs — or a combination.
  </Step>

  <Step title="Browse the Library">
    Start with prebuilt evaluators from the Library. Customize thresholds and prompts to match your quality standards.
  </Step>

  <Step title="Test in Playground">
    Validate evaluators against sample data before deploying to production evaluations.
  </Step>

  <Step title="Attach to Evaluations">
    Connect evaluators to [evaluations](/Evaluations/text-evaluations/Datasets) and map variables to your data fields.
  </Step>

  <Step title="Run and Analyze">
    Execute evaluations and analyze results in [Test Runs](/Evaluations/TestRuns). Iterate on prompts and thresholds.
  </Step>
</Steps>

## Related

* [Text Evaluators](/Evaluators/text-evaluators) - LLM-as-Judge, code, and rule-based text evaluators
* [Voice Evaluators](/Evaluators/voice-evaluators) - TTS, STT, and conversational evaluators
* [Image Evaluators](/Evaluators/image-evaluators) - Multimodal and rule-based image evaluators
* [Evaluation Overview](/Evaluations/Evaluation-overview) - Evaluations, test runs, and the evaluation framework
* [Quick Start: Evaluation](/quick-start/QuickStart_Evals) - Get started with evaluations in minutes