I built a workflow that automatically extracts invoice data from scanned PDFs and sends it to Google Sheets. Here’s how it works: • Sarvam Vision performs OCR on scanned invoices • Gemini extracts structured invoice fields • n8n pushes the cleaned data into Google Sheets So the pipeline becomes: 📄 Scanned Invoice → 🤖 AI extraction → 📊 Structured spreadsheet data This can be useful for automating: - Vendor invoice processing - Expense reimbursements - Bulk document intake - Finance / ops workflows And the best part — it's free to use on n8n. 🔗 Template link: https://n8n.io/workflows/13779 If anyone here is working on document automation or invoice processing, would love to hear how you're solving it. Happy to answer questions about the workflow as well 🙌

While exploring the latest updates in the AI agent ecosystem, I realized many people mix up these three concepts: • Prompting • MCP Servers • Claude Skills But they actually solve very different problems when building AI systems. A simple way to think about it: Prompting → tells the AI what you want MCP → gives the AI tools to access data Skills → define the workflow or SOP So instead of relying only on prompts, modern AI agents combine all three: Prompt → triggers the task MCP → retrieves the data Skills → execute the workflow That’s when an LLM moves beyond being just a chatbot and starts acting like a real AI teammate. Curious — are you experimenting with Claude Skills or MCP servers yet? Would love to hear what you're building. 🚀

If you’re building AI automations, you’ve probably faced this problem: “What if the agent sends the wrong email?” “What if it refunds the wrong amount?” “What if it writes bad data to production?” That hesitation is real. With the new Human-in-the-Loop (HITL) features in n8n v2.5+, we finally have a clean native solution. Here’s what this unlocks 👇 1️⃣ Tool Approvals Your AI pauses before executing sensitive actions. Refunds. Emails. Database writes. You approve → it runs. No approval → no action. 2️⃣ Send Approvals Where You Already Work You can route approval requests to: Slack | Microsoft Teams | Discord | Telegram | WhatsApp | Gmail | Outlook No dashboard hopping. 3️⃣ See Exactly What the AI Is About to Do Not just “Approve action?”You see: - The drafted email - The refund amount - The exact payload So approvals are informed, not blind. 4️⃣ Multi-turn Agent Conversations Agents can now: - Pause - Ask follow-up questions - Wait for clarification - Continue based on your response This makes workflows feel collaborative instead of robotic. The interesting part? They’re exploring editable parameters during review — meaning you’ll be able to tweak the AI output before approving it. That’s huge for real-world deployments. Curious: For those building AI agents here —Are you already using Human-in-the-Loop in production? Or are you still fully autonomous? Would love to hear real setups 👇

Just built an end-to-end document automation workflow using Sarvam Vision + n8n. Goal was simple: Take messy, low-quality scanned documents and turn them into structured, machine-readable data -fully automated. What it does: • Upload document • Run OCR using Sarvam Vision • Extract structured data • Use AI to pull key fields • Automatically update a sheet No manual cleanup. No copy-paste. Everything runs automatically. The interesting part? Sarvam Vision doesn’t just return raw OCR text — it returns structured layout blocks. That makes downstream automation much more reliable. This kind of setup can be used for: - Healthcare forms - KYC processing - Insurance claims - Any document-heavy workflow If anyone here is building with n8n + OCR + LLMs, happy to share the workflow.

This weekend I experimented with a different way to build RAG. Instead of the typical setup: - Generate embeddings - Store in Pinecone / Supabase - Manage vector DB infra - Handle indexing + costs I tested Gemini File Search API directly inside n8n. And honestly… it simplified the entire pipeline. 🔧 What I Actually Built Inside n8n, I used just 4 HTTP requests: 1. Create a file store 2. Upload a document 3. Move the file into the store 4. Query Gemini That’s it. Gemini handled: - Chunking - Embeddings - Indexing - Retrieval No external vector database.No embedding model setup. 💰 Why This Is Interesting - Storage is free - No hourly DB cost - Indexing is $0.15 per 1M tokens For small projects, internal tools, or MVPs — this is extremely cost-efficient. ⚠️ Important Limitations I Noticed This is not magic. - No automatic version control (re-upload = duplicate data) - Chunk-based retrieval struggles with full-document reasoning - OCR works, but messy documents still need preprocessing - Data is processed on Google servers (privacy considerations apply) So architecture thinking still matters. My Take For: - Internal AI assistants - Automation workflows - Startup prototypes - Personal tools This is a powerful alternative to traditional vector DB setups. I wouldn’t blindly replace enterprise-grade systems yet — but for builders, this is very interesting. If anyone here is experimenting with Gemini File Search or building RAG in n8n, I’d love to compare notes 👇 Happy to share the workflow structure if there’s interest.