Wanted to share what I’ve been building recently. Four n8n workflows, each one tackles a completely different problem. I tried to make them as close to production-ready as I could - proper error handling, retries, real data flowing through. Not just demos.
Here’s what I ended up with:
Lead Intelligence System - 16 nodes, 3 AI calls
Started as an experiment to see how much I can squeeze into a single webhook workflow. A lead comes in, JavaScript validates it against 20 rules and scores it 0-100 across 12 factors. Then the interesting part - three sequential Groq API calls, each building on the previous one. First figures out what the person actually wants, second looks up their company, third writes them a personalized message. Switch node at the end routes by priority - P0 through P3, each with its own SLA. The whole thing takes about 8 seconds and returns a JSON response while dropping a full report in Telegram.
The tricky part was chaining three AI calls where each prompt depends on the previous output. Ended up using different temperatures for each - 0.2 for intent, 0.3 for research, 0.7 for writing. Made a big difference.
Market Monitor - 16 nodes, fully autonomous
Got tired of manually checking tech news every morning. Built this to do it for me. Schedule Trigger fires every 12 hours, grabs articles from HackerNews, TechCrunch, and The Verge through RSS. The part I’m most proud of is the NLP scoring engine - about 60 lines of JavaScript, five keyword groups with different weights. A title match for “zapier” counts way more than “automation” buried in the tenth paragraph. Fuzzy deduplication catches the same story across different outlets. Top articles go to Groq for trend analysis. I get a clean briefing in Telegram, and if a direct competitor gets mentioned - separate alert.
Haven’t touched the code in weeks, it just runs.
Support Bot - 15 nodes, real-time Telegram bot
The most interactive project. An actual Telegram bot that responds to customers in under 4 seconds. Parser pulls metadata from each message, Switch splits traffic three ways - commands, feedback, and support requests. Support messages go through two AI calls - classification and response generation. Classification covers 7 categories, 4 urgency levels, and sentiment detection. Then a SLA matrix with 28 combinations decides the deadline - billing plus critical equals 30 minutes, feature request plus low equals 24 hours. Bot answers the customer and simultaneously drops a full ticket to the internal team.
One thing I’m happy with - using temperature 0.1 for classification (need consistency) and 0.6 for responses (need it to sound human). Small detail but it makes a real difference.
Company Research Pipeline - 15 nodes, multi-API enrichment
Probably the most technically fun one. Give it any domain and it builds a profile from public data. Three Google DNS API calls - MX records show the email provider, TXT records reveal which SaaS tools they use (found 14 services on HubSpot including Stripe, Atlassian, and Salesforce), A records identify hosting from IP ranges. Then HTTP GET grabs the homepage and JavaScript scans for 22 technology signatures in the HTML. Everything merges, gets a tech score from 0 to 100, and Groq writes a company profile with a sales recommendation. HubSpot scored 90/100 - which honestly sounds about right.
Fun fact: originally built the DNS calls in parallel but n8n Cloud couldn’t reliably merge three branches. Switched to sequential - adds maybe 200ms but works every single time. Worth the tradeoff.
All four are on GitHub with full docs and screenshots: GitHub - penkayone/n8n-automation-portfolio: Production-grade n8n workflow automations — AI lead intelligence, competitive market monitoring, customer support bot, multi-API company research. 4 enterprise workflows, 62 nodes, 9 AI calls, 40+ tech signatures. · GitHub
Happy to answer any questions about the builds. Also open for hire if anyone needs help with n8n - Telegram @antongoloskokov