Hey everyone 👋 I have got an idea midnight. Here’s a quick breakdown of the idea or the system I am working on right now: The system checks 10 data sources(have plan to add more sources if I get some solid output) every 24 hours using specific keywords: - Reddit, Facebook, Twitter (X) - Google Web search with advanced queries - Niche specific forums Then I use an LLM (like DeepSeek R1 - free) to verify: - Is the person actually looking for help or need the product/services? - Is it a real need, not someone giving advice or promoting themselves? This helps me filter out low-quality stuff and only focus on real or high intent leads. After that, the system extract anything useful: - Username, company name, or website (if mentioned) - Then send that to Clay (or a similar tool) for enrichment to get emails, phone numbers, or business info. If everything checks out, I send the lead to Airtable or Google Sheets — ready to be contacted. Right now I’m starting with 10 sources, but I plan to scale to more platforms once I have a solid base working. I'll use n8n for now to build the workflow, but move to custom code if it works well in the future. Want some answers to my questions: Would love to hear from someone here who built something similar or worked on a lead gen system before? Am I missing anything obvious that could make this better? Does this seem like a well-thought-out system? Or am I missing something major (or maybe even overcomplicating it)? What would you add if you were building something like this? Is this system architecture scalable? P.S. The goal is to generate 5–6 high-intent leads per day (could be more as the system improves) by monitoring multiple specific data sources and filtering out noise automatically.

Hey everyone, I’ve been working on a simple AI workflow using n8n and DeepSeek R1 (the free model), and I'd definitely need some help. Here’s what I’m trying to do: - A user submits a YouTube video link - I use Apify to transcribe it - Then I send the transcript to the LLM - The goal is to generate a LinkedIn-style post based on the video content It worked once. One of the early posts actually came out great, sharp, and on point. I was happy. Then I tried a different video… and everything fell apart. Now I’m getting posts that are vague, generic, or completely miss the point. It’s been over a week of trial and error. I’ve rewritten the system prompt like 30 times - changing wording, tone, structure, nothing sticks. Sometimes I get decent output. But most of the time, it’s way off. So I’m stuck, and I’m starting to wonder if it’s the way I’m prompting or if this model just isn’t capable for this kind of task. If anyone here has built a content-writing agent that works reliably or in production right now, I’d love to know: - How do you write a system prompt that holds up across different inputs? - Is there a framework or method that helped you make it consistent? - Is switching to GPT-4, Claude, or another paid model would solve this? - And how do you actually test/improve these prompts without losing your mind? Not looking for theory or general AI advice, just real-world experiences from people who’ve actually built something like this and made it work. Would really appreciate any input. Thanks 🙏