Discourse community feedback tracker with AI classification and deduplication
Build a workflow that captures community feedback, bugs, and feature requests posted in our Discourse community, deduplicates them into tracked issues, counts unique supporters, and alerts moderators when an issue crosses meaningful support thresholds — giving the community/product team one reviewable record instead of scattered forum threads. Trigger it from a Discourse webhook (post_created events) as the primary entry point. Add an hourly reconciliation job that re-checks recent posts as a safety net in case webhook delivery gaps occur, plus a separate hourly job that scans for issues that just crossed a support-milestone (e.g. 5+ unique supporters since the last alert) and sends a digest. Also include a manual, resumable backfill job that can import up to 90 days of history without generating historical alerts, and a one-time setup step to register the Discourse webhook. Main flow: Receive the Discourse webhook payload (or reconciliation/backfill post batches). Enrich each post: look up the author's email via Discourse, drop internal/employee posts, and resolve the poster's company via your CRM (I used Salesforce, read-only) without retaining the raw email. Route each post: send new/opening topic posts to a stronger AI model for classification (issue summary, category, possible duplicate-issue match); for replies on already-tracked topics, first check for support/agreement keywords cheaply without AI, and only send keyword-qualified replies to a lighter/cheaper AI model to confirm they're a genuine supporting mention rather than noise. Classify each candidate with an AI model (I used Anthropic Claude — swap for your preferred LLM) to decide relevance, category, and duplicate grouping. Persist results in a durable store (I used SQLite on a volume) as issues plus mentions, tracking one supporting mention per unique user per issue, status history, and duplicate suggestions. Notify a team chat channel (I used Slack) immediately when a genuinely new issue/topic is created. Do not notify on individual replies. Separately, send an hourly digest to chat only for issues that just passed a supporter-count milestone, showing topic, description, total/new supporters, and status. Provide a private internal dashboard (web page) where reviewers can filter/export issues, inspect original and supporting mentions, see supporter counts and milestone history, change issue status, and confirm/reject suggested duplicates — plus a companion API endpoint the dashboard calls to fetch data and take actions. Keep all routes private to the internal team space except the Discourse callback, which should use an unguessable-id style webhook route since it's called by an external service. Fail loudly (non-zero exit) on any AI, CRM, chat, or forum API error rather than writing partial records, and only advance milestone tracking after the chat notification actually succeeds, so failures remain retryable. Tools used: Discourse, Anthropic Claude, Salesforce, Slack, SQLite
What this prompt builds
This workflow automates the collection and tracking of product feedback, bug reports, and feature requests from a Discourse community forum. It uses AI to classify posts, deduplicates issues, tracks supporter counts across unique users, and sends Slack notifications when issues cross meaningful thresholds. Community managers gain a centralized dashboard to review, filter, and manage all feedback without manually reading every forum post.
The problem
Community teams managing product forums face a time-consuming challenge: valuable product feedback — bug reports, feature requests, and general product musings — is buried across dozens or hundreds of forum posts, and manually reading every thread to spot and log those opportunities is tedious and error-prone. Teams need a way to automatically surface actionable feedback, deduplicate recurring issues, and track community support without constant manual triage. This workflow solves that by listening to Discourse webhooks, using AI to classify posts as product feedback, deduplicating issues into a single tracked record per topic, counting unique supporters, and notifying moderators via Slack when an issue crosses meaningful support thresholds — giving community and product teams one reviewable, consistent record instead of scattered forum threads.
Solution and impact
The workflow saves community managers significant time on logging, follow-up, and record-keeping by automatically capturing and classifying Discourse posts in real time, enriching them with CRM data to filter out internal accounts, and consolidating duplicate issues into a single tracked record. It counts genuine supporter mentions per issue using lightweight keyword detection plus AI confirmation, sends immediate Slack alerts for new issues and hourly digests for issues crossing supporter milestones, and provides a secure internal dashboard where managers see all feedback at a glance, change issue status, and export data. Teams gain consistent, visible, centralized tracking of community-sourced product feedback without manual effort.