I'm now a n8n Verified Creator. Been learning AI automation- building real workflows, experimenting, failing, and figuring things out step by step. Every build teaches something new. More builds coming soon 🚀 Check out my profile here: https://n8n.io/creators/divyanshugupta/

Been building quietly for the past few weeks and finally ready to share. I submitted SnapCaption AI to the Flowroom AI App Creation Contest — and honestly, I learned way more than I expected during this build. 🚀 What it does You upload a photo → AI actually looks at it → and in ~10 seconds you get: - 3 platform-optimized captions - A 3-slide Instagram story - 30 ready-to-use hashtags Plus 5 tone styles: Funny, Heartfelt, Poetic, Gen-Z, Professional Works for IG, LinkedIn, Twitter. 🧠 The interesting part — Vision AI This isn’t a generic caption generator. The AI analyzes the actual image — scene, lighting, mood — not just keywords. But getting this right was tricky… - It would hallucinate random stuff (“midnight vibes” on a bright beach pic 😅) - Sometimes added irrelevant context like timestamps - Had to build a validation + retry layer to clean outputs 🛠️ Stack - Flowroom (app infra) - GPT-4o Vision (image + captions) - Vanilla JS - CSS animations Tried to make it feel like a real product, not just a demo. 💡 3 key learnings 1. Vision AI = prompt engineering gameIf your prompt isn’t tight, the output drifts fast. 2. UX matters more than AILoading states, copy buttons, history — spent more time here than on the model. 3. Shipping > perfectHad bugs till the last minute. Real-world fixes hit different vs sandbox builds. 🔗 Try it If you post content regularly, would love your feedback (especially on caption quality across different images) Link: https://flowroom.com/room/3SeEI4 Happy to answer anything about the build 👇

Hey everyone 👋 Just published a new n8n template that solves a problem I kept facing while tracking crypto portfolios. 👉 The problem: - Managing portfolios across multiple chains is messy - You only see balances, not real insights - No clear view of risk or allocation - Everything feels manual and time-consuming 👉 What this template does: • Tracks wallet balances across chains (ETH, Polygon & more via QuickNode) • Converts everything into USD • Calculates portfolio allocation (%) • Uses AI to generate insights, risk & suggestions • Analyzes gas fees across chains • Sends a clean report directly to Slack • Can run daily (set it once and forget it) 👉 Think of it as your AI crypto portfolio assistant Would love to hear your feedback or ideas to improve this 🙌 🔗 Free template: https://n8n.io/workflows/15042

Just completed my submission for the“Build the Ultimate Web Crawler Agent with Firecrawl” challenge 💡 What the challenge is about The idea is simple (but powerful): 👉 Use Firecrawl + AI Agent in n8n 👉 Crawl the web 👉 Enrich the data 👉 Turn it into something actually useful Not just scraping… but building a decision-making agent 🧠 What I built (Case: Aisha – Package Evaluator) I built an agent that answers: 👉 “Should I use this npm package or not?” Instead of manually checking: - npm downloads - GitHub activity - issues - docs 👉 The agent does everything automatically and sends a ready-to-use report ⚙️ How it works • Firecrawl → finds npm + GitHub URLs dynamically • GitHub API → stars, issues, last commit • npm API → weekly downloads • AI Agent → generates insights + recommendation • Slack → clean output for decision-making 📊 Output (this is the cool part) Instead of raw data, it gives: • Risk Score (Low / Medium / High) • Adoption Level (Very popular / Niche) • Issue Health• Alternatives with trade-offs • Final recommendation → Use / Consider / Avoid 👉 Basically… a mini tech decision engine 😅 Challenges I faced • Scraping didn’t work for JS-rendered data (npm downloads ) • AI-only approach was slow and inconsistent • Getting correct GitHub repo dynamically was tricky • Handling invalid packages / edge cases 🔑 Biggest takeaway 👉 The real magic was combining: Firecrawl (discovery) + APIs (accuracy) + AI (reasoning) 🤔 Curious If you had this tool… 👉 Would you actually use it before picking a library? Drop a package name below 👇I’ll run it through the agent 😄

If you're building AI agents in n8n, one thing becomes clear very quickly: 👉 Power is easy 👉 Control is hard That’s where Guardrail nodes come in. They help you protect your workflows from sensitive data leaks, prompt injections, and unpredictable AI outputs — without adding complex logic. 🧹 1. Sanitize Text (No AI Required) Before sending anything to an LLM: • Mask PII (phone numbers, sensitive data) • Hide API keys & secrets • Clean unwanted URLs This ensures only safe, controlled input reaches your AI. 🤖 2. Check Text for Violations (AI-Powered) Using OpenRouter: • Detect jailbreak & prompt injection attempts • Keep responses aligned to your use case • Filter NSFW or unwanted content • Add custom rules (prompts / regex) ⚡ Why this is powerful You can stack multiple guardrails in a single node and define actions: ✔ Pass → continue workflow ❌ Fail → trigger alert / stop execution 💡 Real takeaway Most people focus on building AI agentsVery few focus on making them safe & reliable Guardrails are what turn your workflows from:→ experiments→ into production-ready systems Curious — how are you handling safety in your AI workflows right now?