Built Dwadie — an AI B2B Signal Engine using n8n + Gemini

Built with n8n
, , , , , , ,
1

Hi everyone,

I recently completed an AI automation project using n8n and wanted to share it with the community.

Dwadie is an AI-powered B2B Signal Engine that monitors business signals, identifies potential opportunities, enriches company information, and generates personalized outreach.

The workflow includes:

• Business signal discovery
• LLM-based buying signal analysis
• Lead scoring
• Company enrichment
• Decision maker research
• Personalized outreach generation
• Google Sheets CRM updates
• Telegram notifications

Tech stack:

  • n8n
  • Google Gemini
  • Google Sheets
  • Telegram Bot API
  • RSS feeds

I’m currently exploring more advanced AI agent workflows and looking forward to learning from the community.

Would love feedback from other n8n builders.

final-output
final-output1775×922 173 KB

architecture
architecture1750×312 58.3 KB

2 curtidas
2

Nice build.

The part I find most interesting is not only the enrichment or outreach generation, but how you keep the signal quality reliable over time.

How are you handling duplicate signals, stale company data, and false positives before a lead is scored or outreach is generated?

I’m also curious whether the scoring is based on a deterministic rules layer combined with Gemini, or whether the model makes most of the decision. A human approval step before outreach would also be useful to understand.

The RSS → analysis → enrichment → CRM → Telegram flow is practical and easy to follow.

Thanks for sharing.

3 curtidas
3

Thanks, I appreciate the thoughtful feedback.

At the moment, duplicate detection is handled by checking existing records in the CRM before a lead is processed further. For stale data, the workflow currently enriches companies using the latest available information before generating outreach. False positives are reduced by having Gemini first classify whether a news item is a genuine buying signal and assign a confidence score.

The scoring is primarily AI-driven at this stage, with structured prompts guiding Gemini’s decision rather than a separate deterministic rules engine. I see a hybrid approach (AI + business rules) as the next evolution for improving consistency.

The current version also keeps a human in the loop. Outreach is generated automatically but reviewed before it’s actually sent. I prefer that approach, especially for high-value B2B sales.

Really appreciate the questions, they’re exactly the kind of production considerations I’m looking to build on.