Been running this for about 2 months now and it's saved me probably 2-3 hours every single week. The problem I had: I was getting inquiries from Freelancer and email, spending 20-30 min per lead just figuring out if they were worth talking to. Most weren't. Here's what the n8n flow does now: 1. New inquiry comes in (webhook from Freelancer / email parser) 2. 2. AI node reads the message and scores it: budget range, project complexity, timeline, red flags 3. 3. If score < 6/10 → sends a templated "not a fit" response automatically 4. 4. If score ≥ 6/10 → extracts key details, creates a summary card, sends me a Telegram notification with the brief + suggested opening line 5. 5. I get a clean message: "New qualified lead. Budget: €800-1200. Need: n8n automation for real estate CRM. Timeline: 2 weeks. Suggested reply: [draft]" Total time from inquiry to me seeing it: ~45 seconds. My actual decision time: 2-3 minutes max. The whole thing runs on Oracle Cloud Free Tier. €0/month infrastructure. What I used: - n8n (self-hosted) - - OpenAI node for scoring + summarizing - - Telegram node for notifications - - Simple IF node for routing Happy to share the workflow JSON if anyone wants to build something similar. Drop a comment.

Built this for a real estate client a few months back and it's still running 24/7 without issues. The problem: they were manually checking OLX, Storia and IMMOFLUX every morning for new listings in their target area. By the time they called, half the good ones were gone. The solution: a single n8n workflow that scrapes all 3 platforms every 15 minutes, deduplicates against a Postgres database, and pushes new listings instantly to Telegram with price, location, photos and a direct link. The stack: - n8n (self-hosted on Oracle Cloud Free Tier — yes, completely free) - - HTTP Request nodes with custom headers to bypass basic bot detection - - Postgres for deduplication (IF EXISTS check on listing ID) - - Telegram Bot API for instant notifications - - Cron trigger every 15 minutes The part that took the most time wasn't the scraping — it was the deduplication logic. You don't want the same listing pushed 4 times just because the price changed slightly. My solution: store a hash of (listing_id + price + location). If any of those change, it's treated as a new/updated listing and pushed again with an "UPDATED" flag. Total hosting cost: €0/month Response time from new listing → Telegram notification: under 2 minutes If anyone is building something similar or has questions about the dedup logic, happy to share more details.

Most people pay €20–50/month for a VPS just to keep n8n running. I've been doing it for free for months now. The secret: Oracle Cloud Always Free tier. What you actually get for free (forever): - 4 Arm-based CPUs + 24GB RAM - - Fixed public IP - - Block storage - - No surprise charges My stack on it: → n8n in Docker (docker-compose) → PostgreSQL as the database → Nginx as reverse proxy → SSL via Let's Encrypt (certbot) What it's running right now: → Real estate listing monitor — checks OLX every 15 min → New listing → instant Telegram notification → Lead data → Google Sheets auto-update → Claude API for lead qualification scoring Uptime: 99%+. Cost: €0. The setup takes maybe 2–3 hours if you have basic Linux knowledge. The Oracle free tier doesn't expire — it's not a trial. I can share the docker-compose.yml I use if anyone wants it. Drop a comment below and I'll post it 👇 Anyone else running their workflows on Oracle or another free cloud? Curious what stacks people are using here.

Built something last week that changed how I handle inbound leads for my automation services. The problem: getting DMs and form submissions from people interested in n8n automation, but 80% were a bad fit — wrong budget, wrong use case, or just learning. The solution: n8n workflow that intercepts Typeform submissions, runs them through Claude API with a scoring prompt, and routes: - Score 8-10 → calendar link sent automatically - - Score 5-7 → added to nurture sequence in Airtable - - Score 1-4 → gets educational resources, no sales call booked Claude evaluates: budget signals, use case complexity, tech readiness, intent clarity. Results after 7 days: → 40% fewer calls scheduled → 100% of booked calls were actually qualified → Zero time on "I just want to learn n8n" requests Runs on Oracle Free Tier, €0/month hosting. Claude API: ~€3-4 for the week. The real insight: LLMs read between the lines better than any dropdown form field. Tone + specificity + context = way more signal than "what's your budget?". Anyone else routing leads through an LLM inside n8n? Curious what prompts work for you.

Când am început cu n8n, puneam totul într-un singur workflow — 25-30 noduri, totul conectat, de sus până jos. Părea logic la momentul respectiv. Dar când cădea un nod undeva la mijloc, trebuia să iau totul de la capăt. Debugging-ul era un coșmar. Am schimbat abordarea complet: max 7-8 noduri per workflow, restul se apelează prin "Execute Workflow" node. Ce câștig cu asta: ✅ Error handling specific pe fiecare modul ✅ Pot testa fiecare parte separat ✅ Când ceva pică, știu exact unde și de ce Concret la mine: flow-ul de monitorizare OLX are 3 sub-workflows separate — trigger, procesare, notificare Telegram. Dacă Telegram are o problemă, flow-ul de procesare continuă să ruleze fără să se oprească tot. 6 luni pe Oracle Cloud Free Tier, 0 €/lună, 94% uptime la rularea simultană a 11 workflows. Voi cum vă structurați workflows-urile mai mari? Un singur flow sau le fragmentați?