Inspiration

SexismAI was inspired by the need to make online spaces safer for women. Women-focused Reddit communities often deal with misogyny, hostile language, and targeted harassment, and moderators need faster tools to spot harmful content before it damages the conversation.

What it does

SexismAI helps protect women by scanning Reddit posts and comments for misogyny, sexism, hostile language, toxicity, hate, and harassment. It gives moderators a review form with scores, a reason, and recommended actions like remove, lock, approve, or ignore.

How we built it

We built SexismAI as a Reddit Devvit app using native moderator menu actions, Devvit settings, Reddit API access, Redis for recent scan records, and Gemini for real AI classification. Moderators can scan live Reddit content directly from post, comment, or subreddit menus.

Challenges we ran into

The hardest part was working around Devvit’s platform limits for outbound AI provider requests. Some providers were blocked by domain policy, and Gemini quota/rate limits had to be handled clearly. We also had to remove earlier backend assumptions and make the app work fully inside Reddit’s Devvit runtime.

Accomplishments that we're proud of

We built a real Reddit moderation workflow instead of a mock demo. SexismAI runs inside Reddit, fetches real posts and comments, stores recent scans, and gives moderators practical review actions. We are proud that the tool is focused specifically on protecting women-centered communities.

What we learned

We learned how Devvit handles app installation, settings, moderator menus, Redis storage, and HTTP fetch permissions. We also learned that moderation tools need clear failure messages, safe defaults, and human review instead of pretending AI decisions are perfect.

What's next for SexismAI

Next, we want to improve classification quality, add better moderator analytics, support automatic submit triggers once stable, expand configuration for community-specific rules, and build stronger reporting for repeated misogyny and harassment patterns.

Built With

Share this project:

Updates