Righting Sentences AI - Release Notes-2.1

Version: 2.1 Status: Active URL: Launch Application

1. Overview

Righting Sentences AI is a web-based editorial application designed to enforce higher standards of clarity and precision in generative AI outputs. It serves as a specialized program to avoid robotic and repetitive patterns common in responses of commercially available, foundational large language models (LLMs).

This is the first version of the application to have documentation.

2. Purpose

The primary function of Righting Sentences AI is to "humanize" machine-generated text by forcing it to use a larger vocabulary.

Most foundational AI models default to using a specific set of words that summarize complex ideas or smooth over transitions too quickly, and without adding meaning. Many of these words are also overused in the American English vocabulary. Righting Sentences AI prohibits the model from generating these words. By removing the model's ability to rely on these shortcuts, the application forces the AI to:

• Explain concepts clearly rather than summarizing them
• Adopt a sentence structure that aligns more closely with human writing styles
• Avoid the cliches typical of AI-generated content

Note: The application operates strictly as a stylistic filter. It does not alter the context of your prompt, nor does it inject new topics, opinions, or hallucinations into the output. It simply compels the model to express your original ideas using a more rigorous vocabulary.

3. User guide

Interface design

The user interface is modeled after industry-standard generative AI playgrounds. For GenAI users, the learning curve will be shallow.

How to use

1. Select a model: Use the dropdown menu to choose your preferred LLM
2. Write a prompt: Type or paste your draft content into the text input area
3. Generate the text: Select the orange Generate button
4. Review: the output The optimized text will appear in the Result panel

Copying the result

To export your result, select the gray Copy button located in the upper right-hand corner of the output box. The button label will dynamically change to "Copied!" to confirm.

Visual themes (dark/light mode)

The application does not have a manual toggle for choosing between dark and light mode. Instead, it respects your operating system's settings.

• If your computer is set to dark mode, the UI will render in dark grays and charcoal
• If your computer is set to light mode, the UI will render in white and light gray
• To change the appearance, please adjust your device's system settings

4. Technical specifications

Token limits & capacity

The application can process 8,921 tokens per request, which is about 4,000 words.

• Standard usage: GPT-4o and Claude Sonnet 4.5 operate efficiently within this limit
• High efficiency: Gemini 3 Flash is optimized to reduce token consumption for longer drafts
• High fidelity: OpenAI's GPT-5.2 Pro is more intensive and will consume tokens faster

Supported models

The following models are integrated into the application:

• OpenAI GPT-4o
• OpenAI GPT-5.2-Pro
• Google Gemini 3 Flash
• Anthropic Claude Sonnet 4.5

5. Technology stack

Righting Sentences AI uses the following technologies:

• Backend: Python 3.11-slim, Flask, LLM Lite
• Frontend: HTML5, CSS3, JavaScript
• Architecture: Docker, Google CloudRun
• Development tools: Visual Studio Code, Sublime Text, Google Gemini 3

6. Future improvements

• Perform a security review
• Replace Flask with Fast API
• Consider an update to the UI for more capabilities
• Add the option to refresh the text input and canvas fields
• Add file upload capabilities
• Consider integrations and connections