README.md

---
title: AI Humanizer
emoji: "\U0001F9E0"
colorFrom: pink
colorTo: yellow
sdk: gradio
sdk_version: 5.12.0
app_file: app.py
pinned: false
---

# AI Text Humanizer

**3-Agent Sequential Pipeline** for transforming AI-generated text into
natural, human-sounding writing.

## Architecture

```
Input Text -> [Semantic Analyzer] -> [Draft Generator] -> [Humanizer] -> [Verifier] -> Output
                                                              ^                |
                                                              |  (loop if AI)  |
                                                              +----------------+
```

## Models Used (all free HF Inference)

| Agent | Model |
|-------|-------|
| Semantic Analyzer | mistralai/Mistral-7B-Instruct-v0.3 |
| Draft Generator | mistralai/Mistral-7B-Instruct-v0.3 |
| Humanizer | HuggingFaceH4/zephyr-7b-beta |
| Verifier | roberta-base-openai-detector |

## How It Works

1. **Agent 1** extracts topic, tone, audience, and arguments (read-only analysis)
2. **Agent 2** rewrites text naturally while preserving 100% of facts
3. **Agent 3** injects human imperfections: contractions, fillers, hedging, varied sentence lengths
4. **Verifier** checks with AI detector -- loops back to Agent 3 if flagged (max 3x)