ModMind Pro

• 

  risk output

What inspired us

Every Reddit moderator faces the same invisible tax: before making any decision, they manually open the user's profile, scroll through history, check modmail for prior warnings, and cross-reference past bans. This takes 3–8 minutes per case. A subreddit with 50 daily reports loses 4–6 hours of mod time just gathering context — before a single decision is made. We built ModMind Pro to eliminate this entirely.

What it does

ModMind Pro is a moderation intelligence layer that works silently in the background. The moment a post or comment is submitted to a subreddit, ModMind Pro instantly fetches the author's account age, karma, and full violation history, then calculates a risk score from 0–100. By the time a mod opens the queue, everything is already there.

Three core systems running simultaneously:

1. Proactive Context Engine Triggers on PostSubmit and CommentSubmit — not after a mod action. Context is pre-built and cached in Redis before any mod opens the queue. Zero waiting. Zero manual lookup.

2. Risk Scoring Algorithm Every user gets a risk score calculated from:

• Account age (new accounts score higher)
• Karma (low or negative karma increases risk)
  
• Violation history (each prior removal adds 15 points)
• Maximum score: 100

$$\text{Risk} = \text{AgePenalty} + \text{KarmaPenalty} + (\text{Violations} \times 15)$$

Scores map to three levels: 🟢 LOW (0–34), 🟡 MEDIUM (35–69), 🔴 HIGH (70–100)

3. Pattern Detection & Coordination Alerts ModMind Pro learns from every decision. Three detectors run on every new submission:

• Repeat Offender Detector: flags users with 2+ violations in 7 days at up to 95% confidence
• Coordination Detector: flags when 3+ new accounts post within 60 minutes — the primary signal of brigading and spam campaigns
• Pattern Matcher: compares new users against 500 historical decisions. If 70%+ of similar profiles were removed, fires a pattern alert

How we built it

Built entirely on Reddit's Devvit platform using TypeScript. Four core files:

• stateManager.ts — Redis schema, risk score formula, decision logger, violation tracker, alert system
• contextBuilder.ts — parallel Reddit API fetching with graceful fallbacks for deleted/suspended accounts
• detector.ts — three independent detection algorithms running in parallel via Promise.all
• main.ts — Devvit triggers wiring everything together

Every decision is logged permanently to Redis, building institutional memory that makes the system smarter over time. A subreddit that has been running ModMind Pro for 30 days has 500 historical decisions powering its pattern detection — something no new mod team has today.

Challenges we faced

The proactive trigger problem: Devvit's ModAction trigger fires after a mod acts, not before. We solved this by switching to PostSubmit and CommentSubmit triggers, so context is built the moment content arrives — before any human sees it.

The authorId problem: PostSubmit events return authorId, not authorName. We had to resolve the username via getUserById before building context, adding an extra API call to the pipeline.

Graceful degradation: The tool must work even when the Reddit API is slow or a user is deleted/suspended. Every API call is wrapped in try/catch with meaningful fallbacks — a deleted user gets a risk score of 60 (unknown = elevated risk), not a crash.

What we learned

The most valuable insight: moderation problems are fundamentally information problems. Mods don't make bad decisions because they lack judgment — they make inconsistent decisions because they have inconsistent information. ModMind Pro solves the information problem first, and everything else follows.

What's next

• AI-powered content classification using Anthropic API
• Cross-subreddit bad actor sharing between opted-in communities
  
• Mod team analytics dashboard showing queue health over time
• Weekly digest reports sent to mod teams automatically

Built With

• api
• devvit
• node.js
• reddit
• redis
• typescript