GPT Humanizer Overview

This project exists to explore human-centered control over LLM output without turning the system into a policy engine or a brittle ruleset.

The goal is not to censor language, but to:

Shape tone and wording deliberately
Preserve meaning while avoiding specific terms
Create space for “humanizing” transformations downstream

Design priorities

Simplicity over abstraction
Observability over automation
Clear separation between generation, evaluation, and rewriting
Minimal UI that reveals behavior rather than hides it

This is a foundation, not a product.

Future work may include additional rewrite strategies, tone shaping, and documentation-first development, but the system should remain understandable end-to-end.

A lightweight FastAPI service that generates text via an LLM, applies banned-word detection, and optionally rewrites output to preserve meaning while avoiding specified terms. A minimal headless web UI is included for interactive use and inspection of requests and responses.

Features

Plain-text generation endpoint
Optional banned-word filtering with rewrite pass
Temperature and token controls
Headless browser UI for prompt entry and testing
No client-side model logic

Requirements

Python 3.11+
An OpenAI API key

Setup

1. Create and activate a virtual environment 2. Install dependencies from `requirements.txt` 3. Export `OPENAI_API_KEY` in the shell 4. Run the server with Uvicorn

Usage

API endpoint: `POST /generate`
UI: open `http://127.0.0.1:8080/` in a browser
Health check: `GET /healthz`

The `/generate` endpoint returns plain text only.