ModMind-Ai

About the Project

What inspired us

Reddit moderators are some of the most underappreciated people on the internet. They volunteer hours of their time every single day manually reading through posts and comments, trying to keep their communities healthy. We saw this problem firsthand — mod queues piling up, toxic content slipping through, spam bots running rampant — and thought: there has to be a better way. That's what inspired ModMind AI. Not just another moderation tool, but a full intelligence platform that works for moderators instead of adding to their workload.

What we learned

Building on Devvit was a completely new experience. We learned how Reddit's developer platform works from the ground up — how triggers fire on real content, how Redis storage works within the Devvit ecosystem, and how to build a production-grade React dashboard that lives entirely inside Reddit. We also learned a lot about content moderation itself — the patterns, the signals, and the behavioral cues that separate genuine violations from false positives.

How we built it

ModMind AI is built entirely on Devvit — Reddit's native developer platform. The detection engine uses a multi-layer rule-based system combining keyword scoring, regex pattern matching for spam and hate speech, and behavioral signals like excessive caps and character spam. The dashboard is built in React with a clean dark professional UI, and all data is stored in Devvit's built-in Redis store — meaning zero external infrastructure required. Every post and comment submitted to the subreddit is automatically analyzed by the trigger system. Items scoring above the threshold are flagged, scored, explained in plain language, and surfaced to moderators in a prioritized queue.

Challenges we faced

The biggest challenge was working with Devvit's newer web template — the documentation is still catching up with the latest version, so we had to dig deep into the CLI source code to understand the correct config schema. We also had to rethink our architecture when we realized triggers don't fire reliably in playtest mode, which led us to build the Simulate Scan feature as a powerful demo and testing tool. Getting the detection engine to be accurate without any external AI API — purely rule-based — was also a satisfying engineering challenge.

Built With

• devvit
• hono
• node.js
• react
• redis
• typescript