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

# Guardian API Documentation

> Multi-model, multi-label content moderation service

## Welcome to Guardian API

Guardian API is a production-ready content moderation service that combines multiple AI models and rule-based heuristics to provide comprehensive text analysis. Built with FastAPI, it offers a robust, scalable solution for detecting harmful content across multiple categories.

<Card title="Get Started" icon="rocket" href="/quickstart" horizontal>
  Start moderating content in minutes with our quickstart guide.
</Card>

## Key Features

<CardGroup cols={2}>
  <Card title="Multi-Model Architecture" icon="brain">
    Four coordinated models working together: Sexism classifier, toxicity transformer, rule engine, and ensemble aggregator.
  </Card>

  <Card title="Multi-Label Detection" icon="shield-check">
    Detects sexism, toxicity, threats, self-harm, profanity, and other harmful content categories.
  </Card>

  <Card title="Production-Ready" icon="server">
    FastAPI backend with proper error handling, logging, rate limiting, and comprehensive testing.
  </Card>

  <Card title="Developer-Friendly" icon="code">
    OpenAPI documentation, structured JSON responses, and SDKs for Python and JavaScript/TypeScript.
  </Card>
</CardGroup>

## How It Works

Guardian API uses a sophisticated ensemble approach:

1. **Preprocessing**: Text is cleaned and normalized
2. **Model Inference**: Three models analyze the content in parallel:
   * Custom LASSO sexism classifier
   * HuggingFace toxicity transformer
   * Rule-based heuristics engine
3. **Ensemble**: Results are aggregated using weighted fusion and conflict resolution
4. **Response**: Structured JSON with scores, severity levels, and metadata

<Note>
  The ensemble model intelligently combines outputs from all models, with rule-based detections overriding low confidence scores for critical issues like threats and self-harm.
</Note>

## Quick Example

```bash theme={null}
curl -X POST "http://localhost:8000/v1/moderate/text" \
  -H "Content-Type: application/json" \
  -d '{"text": "Your text to moderate here"}'
```

**Response:**

```json theme={null}
{
  "text": "Your text to moderate here",
  "label": {
    "sexism": {"score": 0.12, "severity": "low"},
    "toxicity": {"overall": 0.08},
    "rules": {"slur_detected": false, "threat_detected": false}
  },
  "ensemble": {
    "summary": "likely_safe",
    "score": 0.10,
    "severity": "low"
  },
  "meta": {
    "processing_time_ms": 24,
    "models_used": ["sexism_lasso_v1", "toxic_roberta_v1", "rules_v1"]
  }
}
```

## Next Steps

<CardGroup cols={2}>
  <Card title="Quickstart" icon="play" href="/quickstart">
    Set up and run Guardian API locally
  </Card>

  <Card title="Architecture" icon="sitemap" href="/concepts/architecture">
    Learn about the multi-model architecture
  </Card>

  <Card title="API Reference" icon="book" href="/api-reference/moderate-text">
    Explore the REST API endpoints
  </Card>

  <Card title="SDKs" icon="code" href="/sdks/python">
    Use our Python or JavaScript SDKs
  </Card>
</CardGroup>

## Use Cases

* **Social Media Platforms**: Moderate user comments and posts
* **Chat Applications**: Real-time message filtering
* **Content Platforms**: Screen user-generated content
* **Community Forums**: Maintain healthy discussions
* **Customer Support**: Flag inappropriate messages

## Author

Guardian API is developed and maintained by **Kory Smith**.

Originally created as a CSC 380 (Machine Learning) sexism classification project at the University of Arizona.