Mod Sense

Inspiration

I wanted to help moderators deal with changing spam without maintaining complex AutoModerator rules or sending community data to outside services.

What it does

Mod Sense learns from normal remove, spam, and approve actions, then scores new posts locally to help decide what should be removed or reviewed.

How we built it

I built it as a Devvit app with TypeScript, Devvit Redis, and a small Naive Bayes classifier, so it does not use external AI APIs or off-platform databases.

Challenges we ran into

The hard part was making it useful without external services, especially for cold starts, conservative thresholds, and clear score explanations.

Accomplishments that we're proud of

I’m proud that it fits into existing moderator workflows and keeps the learning data inside Reddit’s Devvit environment.

What I learned

I learned that a simple local model can be enough for reducing repeated moderation work when it is meant to support, not replace, moderators.

What's next for Mod Sense

Next, I want to improve testing, make AutoModerator imports more useful, and give moderators finer control over how the model learns.

Built With

• devvit