ModIAGuard

AI-powered moderation suite for Reddit (Devvit Hackathon).

Comment

Inspiration

Moderating a subreddit means juggling reports, spam, toxic language, and unclear context—all while keeping a clear audit trail. Mod tools are powerful but scattered. We built ModAiGuard so moderators can analyze content, get AI-backed recommendations, and act on the mod queue without leaving Reddit, using Devvit Web.

What it does

ModAiGuard is an AI moderation copilot for subreddit mods:

  • Toxicity Analyzer — scores harmful language (insult, threat, hate) via OpenAI
  • Spam Detector — flags links, repetition, and promo patterns
  • Report Review — triage the real mod queue with AI banners and one-click actions (approve, remove, spam, needs context)
  • Auto-Moderation — process the entire mod queue automatically from a single button
  • Auto-Responder — trigger-based reply rules when content matches
  • Logs, History & Moderations Panel — full audit trail with scores and recommendations
  • Demo Mode — fake posts for judges/demo days (no Reddit API calls)
  • Activity notifications — recent moderation events in the header

Mods open the panel from the subreddit menu: “Abrir ModAiGuard Panel”.

How we built it

We used Devvit Web (Reddit’s app platform) with:

  • Backend: TypeScript + Hono REST API (/api/toxicity, /api/spam, /api/ai/recommend, /api/report-actions, /api/auto-moderation, /api/demo, etc.)
  • Frontend: React 19, React Router, Vite, Tailwind CSS 4
  • Storage: Redis for moderation logs (AI_LOGS_<subreddit>) and demo posts
  • Reddit API: mod queue, approve, remove, mark spam, ignore reports
  • AI: OpenAI gpt-4o-mini for toxicity; heuristic engine for spam; threshold-based recommendation engine

Every API returns { ok: true, data } or { ok: false, error } for consistent error handling.

Challenges we ran into

  • Wiring a full React dashboard inside Devvit’s playtest/simulator iframe
  • Keeping log fields consistent across human actions, auto-moderation, and demo simulations
  • Balancing real auto-moderation (powerful but risky) with Demo Mode (safe for demos)
  • Making risk scores visually clear (green / yellow / red) across many modules

Accomplishments that we're proud of

  • End-to-end flow: analyze → recommend → act → log
  • Real batch auto-moderation on the mod queue
  • Demo Mode so anyone can try the product in under a minute
  • Unified moderation history with toxicity, spam, AI recommendation, and auto flag

What we learned

Devvit is a strong fit for mod-facing tools: rich UI + server + Redis in one stack. Splitting analysis, recommendation, and action made the system easier to test and explain to users—and essential for hackathon demos.

Built With

Share this project:

Updates

Submission history