Original Summary

For several months now I've been developing n8n automation workflows, not as a product, but as project work I'm doing for my clients and some experiments I'm doing on my own. I simply open-sourced all 5 thus decided to share what they look like and what I learned. Here's the lineup: The Full agency operations engine.The entire agency operations engine. Client intake via a web form, invoicing via Stripe, DocuSign NDA, creation of Slack channels, project task generation, status reporting weekly. Client health scores are calculated using a "Pulse Engine" that takes into account email frequency, payment time, number of revisions and margin health. A silent client is a warning sign that an account is at risk of being lost, so you should check before it happens. The most challenging aspect: the calibration of the health scoring. Every new client who has been brought in appears "at risk" if starting to score on day 1, otherwise there is no baseline. Wanted to create a 30 day cold start window, where it just pulls in data, but doesn't alert. A multi-LLM intent radar for LeadGen Engine (95 nodes). Scans Reddit, Twitter, RSS for 'people who are actively stating a problem that I can solve'. Uses a 3-model pipeline: GPT-4o scores intent (0-100), Claude 3.5 Sonnet drafts a contextual reply, GPT-4o-mini is a compliance filter that catches anything that sounds spammy. All things pass via Slack interactive buttons before anybody sees them. The most headache inducing: OpenAI sending back malformed JSON in markdown code blocks. Went so far as to write a JS function node to strip the backticks and attempt to JSON.parse and catch any errors and then fall back to regex extraction. It's annoying, but it does happen more than you would think at volume. RestoFlow (100+ nodes) – Restaurant operations Created this for a restaurant owner who was losing customers due to not receiving phone calls. Inbound calls are answered by AI voice agent (Vapi), bookings and takeout orders are processed to WhatsApp, and automated reminders keep customers from missing appointments. In addition includes loyalty engine and monitoring review sentiment. When the no-show rate was lowered by introducing the 24h + 2h double reminder system with one-click cancel links, it has been significantly reduced. It turns out that the majority of those who don't show up aren't evil, they just forget. ZedProp (60+ nodes) – WhatsApp property management Student accommodation management via WhatsApp. Tenants send text messages in natural language - it translates their intentions (check balance, download lease, report maintenance), and forwards the message to the relevant Airtable lookup or PDF generation pipeline. MedFlow (120+ nodes): Medical practice operations. Appointments, automated pre-visit forms, insurance verification and follow-up post appointments. --- When building 500+ nodes: Error handling cannot be deferred. If you aren't wrapping e

  • 情报分类：商业与市场研究
  • 分类依据：内容涉及商业、投资或市场动态
  • 信息来源：Reddit · SideProject
  • 发布时间：2026/9/20 02:21:39