Generate LinkedIn posts using Telegram, Supabase vector DB and OpenAI RAG

Description

Overview
AI-powered n8n workflow that creates viral LinkedIn posts by learning from successful content. Features two modules: (1) Telegram-based scraper that builds a vector database of viral LinkedIn posts, and (2) Web form that generates optimized posts using multi-agent AI with RAG (Retrieval-Augmented Generation) from your curated viral content library.

Key Capabilities:
Scrapes LinkedIn post content via Telegram bot
Stores posts in Supabase vector database with OpenAI embeddings
3-agent system analyzes hooks, structures outlines, and generates posts
RAG integration retrieves similar viral posts for pattern matching
Auto-publishes to LinkedIn or provides formatted output

How It Works

Module 1: Viral Post Collection (Telegram Bot)

Step 1: URL Validation
User sends LinkedIn post URL to Telegram bot
Workflow validates URL contains "linkedin.com"
Shows typing indicator for better UX

Step 2: Content Scraping
HTTP request fetches post HTML
CSS selector extracts main commentary: [data-test-id="main-feed-activity-card__commentary"]
Handles scraping failures with error messages

Step 3: Vector Storage
Converts post text to OpenAI embeddings (text-embedding-ada-002)
Stores in Supabase linkedin_post table with vector indexing
Sends success confirmation via Telegram

Module 2: AI Post Generation (Web Form)

Stage 1: Hook Analysis Agent
Input**: User-provided hook text
Process**: AI extracts topic, niche/industry, emotional tone, and 3-5 key points
Output**: Structured JSON with analyzed elements
Models**: GPT-4o-mini or Gemini 2.5-flash (dual fallback)

Stage 2: Post Structure Agent
Input**: Analyzed hook data
Process**: Creates 5-section outline (Hook, Problem, Value/Lesson, Solution, CTA)
Output**: Structured framework for final post
Models**: GPT-4o-mini or Gemini 2.5-flash

Stage 3: Post Generator Agent (RAG)
Input**: Post structure + topic
RAG Process**:
Queries Supabase vector store for 5 most similar viral posts
Analyzes patterns: hooks, storytelling, CTAs, engagement metrics
Identifies optimal length, formatting, and emotional triggers
Output**: Complete LinkedIn post applying viral patterns
Models**: GPT-4o-mini or Gemini 2.5-flash with GPT-5-NANO for structured output

Stage 4: Publication
Auto-publishes to LinkedIn via API
Or returns formatted post text for manual posting

How To Use

Setup

1. Configure Supabase Vector Database
Create Supabase project
Create table: linkedin_post with vector column (1536 dimensions for OpenAI embeddings)
Enable vector extension: CREATE EXTENSION vector;
Update credentials in "Upload Document" and "Supabase Vector Store" nodes

2. Set Up Telegram Bot (Module 1)
Create bot via @BotFather
Get bot token and update "On Telegram Message" credentials
Start bot and get your chat ID
Activate workflow

3. Configure OpenAI API
Add API key to "Embeddings" nodes (both modules)
Configure language model credentials (GPT-4o-mini, GPT-5-NANO)

4. Set Up LinkedIn API (Optional for Module 2)
Create LinkedIn app with member permissions
Configure OAuth2 credentials in "Create a post" node
Or remove node to get text output only

5. Access Web Form
Get form URL from "LinkedIn Form" webhook
Bookmark for easy access