Toxic Pulse

AI-powered community health dashboard for Reddit moderators that is built with Devvit + Google Gemini. Generates alerts when toxicity spikes or sentiment drops.

Comment

Toxic Pulse AI

Inspiration

Moderators spend a huge amount of time reacting to individual events: reviewing reports, removing posts, managing queues, and resolving conflicts. But communities are much more than isolated actions. They behave like living systems where patterns emerge over time.

The idea behind Toxic Pulse AI came from a simple question:

What if moderators could understand the health of a community instead of only responding to individual problems?

Across previous work involving AI systems, sentiment analysis, vector search, and intelligent workflows, one recurring observation appeared: the problem is often not a lack of data — it is a lack of visibility into the signals hidden inside that data.

Moderators already have access to:

  • Posts
  • Comments
  • Reports
  • User activity
  • Moderation actions

The challenge becomes finding meaningful patterns quickly enough to take action.

Community Pulse AI was created to transform those raw interactions into insights that moderators can use immediately.

What it does

Toxic Pulse AI continuously monitors subreddit activity and converts community behavior into actionable intelligence.

Features include:

  • Community health scoring
  • Topic clustering
  • Duplicate discussion detection
  • Sentiment analysis
  • Toxicity monitoring
  • Trending topic discovery
  • Smart moderation alerts
  • Actionable moderator recommendations

Example dashboard output:

Community Health: 82/100

Positive engagement +15%
Toxicity −7%

Alerts:

Duplicate RTX 7090 discussions detected
Heated discussion trend emerging

Recommendations:

- Create megathread
- Monitor discussion activity

Rather than requiring moderators to manually discover trends, Toxic Pulse AI proactively surfaces them.

How we built it

Toxic Pulse AI was built using Reddit's Developer Platform with Devvit and TypeScript.

The system follows an event-driven architecture:

Posts + Comments
        ↓

Devvit Triggers
        ↓

Background Processing
        ↓

Analysis Pipeline
        ↓

Community Insights
        ↓

Dashboard + Alerts

The application captures posts and comments through Devvit triggers and sends them through an analysis pipeline.

The pipeline performs:

  • Sentiment analysis
  • Toxicity detection
  • Topic clustering
  • Duplicate discussion detection
  • Trend analysis

To simplify community activity into a single high-level metric, we introduced a community health score:

$$ HealthScore = 0.35(Engagement) +

0.30(Sentiment)

0.25(Toxicity)

0.10(Reports) $$

with normalization:

$$ 0 \le HealthScore \le 100 $$

This allows moderators to quickly understand overall community conditions without manually interpreting multiple metrics.

State and metrics are stored through Devvit storage and updated using scheduled background jobs.

Challenges we ran into

Detecting duplicate discussions

People can discuss the same topic using very different wording.

For example:

RTX 7090 overheating problem

and:

Why is my graphics card getting hot after the latest update?

represent nearly identical discussions while sharing few exact keywords. This required moving beyond basic keyword matching toward semantic similarity and clustering approaches.

Avoiding alert fatigue

An analytics system that generates too many alerts becomes difficult to use. Finding the balance between:

  • meaningful warnings
  • normal community activity

Required careful threshold tuning.

Accomplishments that we're proud of

We're proud of building something that shifts moderation from reactive workflows toward proactive intelligence.

Some highlights include:

  • Creating a measurable community health score
  • Detecting duplicate discussions automatically
  • Building a system that generates actionable recommendations instead of raw statistics
  • Creating a complete workflow from collection → analysis → dashboard insights
  • Designing the project around assisting moderators rather than replacing them

What's next for Toxic Pulse AI

Future versions of Toxic Pulse AI could include:

  • Brigading and coordinated activity detection
  • AI-generated megathreads
  • Moderator workload prediction
  • Retrieval-based moderator memory
  • Community growth forecasting
  • Cross-subreddit trend analysis
  • Automated community summaries
  • Personalized moderation recommendations

Built With

Share this project:

Updates