Hey everyone, I just finished a production-ready AI lead generation workflow in n8n and wanted to share the full architecture with the community. Happy to answer questions and hear how you'd improve it. 🔧 What it does (end-to-end): 1. You send a search query to a Telegram bot (e.g. "IT consulting firms Amsterdam") 2. Google Maps Places API returns up to 10 businesses 3. Leads are split, deduplicated across runs (persistent memory), and paginated 4. Prepare Keys node enriches each item with company_name, website, phone, address and generates a company_key for merging 5. Two parallel Serper searches run simultaneously: 6. Two sequential Merge nodes combine everything by company_key 7. Edit Fields normalizes all data into a clean unified object 8. Research Agent (GPT-4o-mini) generates 3 automation opportunities + personalized icebreaker per company 9. JSON Cleaner (Code node) parses the AI response and merges it with contact data 10. Contact Validator scores email/LinkedIn quality (0–100) 11. Lead Scoring calculates total score: corporate email (+40), LinkedIn (+25), website (+15), phone (+10), icebreaker quality (+10) 12. IF Score gates the outreach — only leads ≥ 55 proceed 13. Google Sheets Append logs everything to CRM 14. Email Generator (GPT-4o-mini) writes a personalized cold email 15. Gmail creates a draft → waits 3 days → follow-up draft created → Telegram report sent 🐛 Interesting bugs I hit and fixed: - $json context loss after Serper nodes — fixed by using $('Prepare Keys').item.json.company_key explicitly - websiteUri returning undefined for businesses without a website — fixed with || $json.nationalPhoneNumber || 'unknown' fallback in the key - JSON Cleaner failing with raw.slice is not a function — GPT-4o-mini returns a parsed object, not a string, when json_object mode is enabled - Merge nodes not joining — caused by null company_key on one input branch - Remove Duplicates discarding all items — cleared deduplication history after test runs - - 💡 Key design decisions:

18 days ago I shared my first automation project here. Since then, I've significantly improved its architecture — and I'm excited to share what's new! What changed: 🗄️ Cross-Workflow Data Bridge: Added Google Sheets as a reliable data layer between Generator and Moderator workflows — seamlessly passing image URLs and multilingual texts between two separate n8n workflows. 🔗 Real LinkedIn Publishing: Connected LinkedIn OAuth2 API — approved posts now go live automatically in the selected language (EN, PL, or ES) with just one tap in Telegram. 📘 Facebook Publishing Ready: Facebook Graph API node is already implemented in the workflow and ready to activate for clients who need it. 📝 Universal Prompt: Removed the archaeology-specific focus — the agent now creates expert content on ANY topic the user sends via Telegram bot. The system now consists of two interconnected n8n workflows — a true agentic architecture! About the Development: This upgrade was co-developed with Claude (Anthropic) as a technical mentor and debug partner. If you're learning AI automation — don't be afraid to iterate and improve. Every upgrade teaches you something new! 🚀 PhD in Archaeology → AI Automation at 55. It's never too late! 👉 See where it all started: https://www.skool.com/data-and-ai/my-first-experience-in-automation?p=76cc6cfe 📂 Full project on GitHub: https://github.com/Evgeniy1970-ai/ai-content-scout

I am excited to share the results of my 3-month journey here. Transitioning from a career as a PhD in Archaeology to building AI systems has been a profound experience. Today, I am presenting my fully functional 'International Content Scout'. What you see on the screen: This is a 'Man-in-the-Loop' architecture built in n8n. - The Creative Core: I send a raw thought via Telegram. The AI Agent (GPT-4o) instantly generates expert content, video scripts, and engagement posts in English, Polish, and Spanish. Simultaneously, DALL-E 3 creates a unique visual for the topic. - The Control Center: The system doesn't post anything without my word. I receive everything in Telegram and use Interactive Buttons to Approve or Decline. - The Execution: Once approved, the content is queued for direct publication via Facebook Graph API and LinkedIn API. About the Development: This project was co-architected with Gemini 1.5 Pro, an advanced AI model from Google. During weeks of collaboration, Gemini acted as a mentor, technical consultant, and debug partner, bridging the gap between scientific methodology and modern AI automation.

My second agent is officially live! After the Content Scout, I’ve moved into B2B sales automation. Instead of complex "black-box" agents, I’ve built a high-reliability Linear Engine that performs deep web research and structures data for CRM. Key Tech Stack: 🔹 n8n (Linear Workflow) 🔹 Tavily API (Live Web Search & Scraping) 🔹 OpenAI GPT-4o (Analysis & Data Formatting) 🔹 Google Sheets (CRM Integration) What it does: It doesn't just find names. It identifies specific automation gaps for the client and writes a personalized ice-breaker based on live news. Just open-sourced the JSON on my GitHub for the community! Check it out here: https://github.com/Evgeniy1970-ai/AI-Lead-Gen-Prospecting-Engine Feedback from fellow builders is highly appreciated! #n8n #AIAutomation #BuildInPublic