About Us

What We Do

We empower moderators with state of the art solutions to moderate hate speech and drastically reducing the workload of those that run online communities. Working directly with community leaders, through programs like our Subreddit Anti-Toxicity Initative, we build tailored solutions through feedback and collaboration, helping those that may lack the technical guidance or resources to implemented automation.

We foster healthy, inclusive communities, by helping remove hateful, toxic, and offensive content before it has time to propagate and spread -- cutting hate at its root, and preventing long chains of conflict. Powered by machine learning, we can accurate identify and remove hateful content in a matter of seconds.

We seek to understand hate speech and toxicity online: why it exists, how it propagates, root causes, and how it can be systematically addressed at scale. These goals serve as the driving force behind large-scale research programs, like our Twitter Toxicity Index

Finally, we help build communities, preventing users from disengaging as the result of slurs or attacks, and leaving community owners time to engage with, promote, and improve their online platform.

Why We Do It

Statistics, of course, exist on the widespread of impact of toxicity online. 2 in 5 Americans experienced harrassment online (ADL). 1 in 4 women experience online abuse (Amnesty International). 23% of teens come across racist comment on a regular basis (Common Sense Media).

Yet, good solutions are rare. Human moderators don't scale, and require volunteer to read through thousands of the most disparaging, hateful content online, while many times receiving personal threats in the process. As Facebook, Twitter, and Discord turn to automated means of tacking content moderation online, smaller communities -- those that number in millions of members -- threaten to be left behind, many times being unable to make the significant investment required to implement AI-based filtering.

Other companies offer solutions. Google's Jigsaw has paved the way for a lot of our own work. And yet, there exists a difference between simply the detection of online toxicity through machine learning, and providing actionable data that moderators can implement.

Providing actionable data means recognizing each community is unique. Each community has its own moderation standards, and thus it is imperative that a range of prebuilt solutions are augmented by the ability to develop customized profiles, tailored to an individual use case.

Providing actionable data means recognizing that AI will never be perfect, and moderators need to find a balance between detection and false positives. For us, that means providing high-confidence predictions that can be immediately addressed, but also the option to flag potentially content that could potentially be harmful. This drives us to implement a precision-based, curated approach to data collection and training

Providing actionable data, most importantly, needs to be accessible and cost friendly. Elaborate, complex, and expensive solutions fail to meet the needs of of online communities. Our work is, and will always be, completely free.

Who We Are

Led by Welton, a machine learning and systems engineer, we build solutions and research that scales. Our work, though, would not be possible without the help of dozens of moderators online, who provide guidance, feedback, and contribute directly to our efforts through data, code, and more.


Our infrastructure is funded by our generous donors and sponsors. We'd especially like to thank the following for their in-kind and monetary support:

DigitalOcean and Linode who have provided crucial parts of the infrastructure necessary to run our programs, databases, and more.

The Hack Foundation who provide fiscal sponsorship, allowing us to inherit their non-profit status and making contributions tax-deductible.