A System for Detecting and Improving Dysfunctional and Toxic Language

Description

Thank you for your interest in our project.

Dailogy is a project dedicated to developing tools that detect dysfunctional and toxic language in chat conversations and suggest ways to make communication more respectful and inclusive.

Here are some of the projects I worked on:

• App Development: This project focused on building an app designed to transform dysfunctional and toxic language into more respectful and inclusive dialogue. The primary use case centered around communication between couples or ex-partners who need to stay in touch. By developing tools that detect and reframe harmful language, we aimed to support healthier, more empathetic interactions. Read the full story.

• Synthetic Data Generation: Due to the sensitive and private nature of couple conversations, collecting real examples of toxic language was not feasible. This project tackled that challenge by generating synthetic data that realistically modeled toxic behaviors—such as insults, threats, and manipulation—across a range of relationship dynamics. These synthetic examples were then embedded to support downstream tasks, and also used in advanced prompting techniques, such as dynamic few-shot prompting, to improve large language models’ ability to detect and rephrase harmful language. Read the full story.