Trending Society

The Platform

Trending Society is an end-to-end AI content platform. It processes content from RSS feeds and social media through a multi-stage AI pipeline, and publishes SEO-optimized articles at scale. Every step is observable, every cost is tracked, and the entire system runs autonomously.

Architecture

The platform is a Turborepo monorepo with clear separation of concerns:

• Content Site: Next.js 16 on Vercel. Public reader-facing app at trendingsociety.com with RSS feed, news sitemap, and llms-full.txt for AI Engine Optimization (AEO).
• SaaS Dashboard: Multi-tenant admin built on MakerKit with role-based access control. Article CMS, newsletter composer, social ingestion grid, media library, and cost analytics.
• Pipeline Engine: 95 Trigger.dev tasks across 11 domains (article, social, content, media, cron, sync, AI, editorial, newsletter, video, repurpose). Full Langfuse tracing on every LLM call.
• MCP Server: Cloudflare Workers-based Model Context Protocol server with 34 tool groups across 33 platform integrations. Powers all AI agent workflows.
• Database: Supabase (PostgreSQL) with RLS, multi-tenant isolation, and event-sourced analytics.

Pipeline

The article pipeline is the core engine. For each piece of source content:

1. Ingest: Aggregate from 19+ RSS feeds and social channels via structured data extraction and RSS automation
2. Classify: AI categorization and topic extraction with Gemini
3. Research: Web research enrichment via Tavily for factual grounding
4. Rewrite: Long-form SEO article generation with quality gates (HTML cleanup, citation audit, FAQ schema)
5. Image: Editorial image generation via FLUX on Replicate with vision-enriched prompts
6. Publish: Autonomous publishing with metadata, structured data, and social distribution
7. Newsletter: Automated weekly newsletter assembly and Resend dispatch

Quality & Observability

Every AI generation is traced through Langfuse with cost tracking, quality scoring, and prompt versioning. The platform has processed 10,840+ AI generations with full audit trails. Quality gates catch and fix LLM output issues (hallucinated URLs, malformed HTML, thin content) before any article reaches production.

Scale

• 740+ published articles across multiple content verticals
• 1,440+ ingested sources from social media and RSS
• 12 automated cron schedules handling daily operations
• Full cost analytics with per-article cost tracking across all AI services