Cold Email Outreach AI Agent, own project
Cold Email Outreach AI Agent
Cold leads to warm meetings: AI-powered sales prospecting
The figures
- Leads in databasegestorias across Spain
- 1,852
- Leads researchedwith enriched data
- 510
- Emails sent4-step sequence
- 419
The problem
Manual gestoria prospecting is slow and unscalable. Researching each lead, writing personalized emails, and following up consumes hours that should be spent closing deals.
What I did
4-step pipeline: deep lead research (Google Maps reviews, PageSpeed, SEO, SSL, competitors), AI-personalized email generation, warm-up sequence (10→20→30/day) and webhook tracking (opens/clicks/replies).
The result
From 0 to 1,852 leads and 419 emails sent in the first weeks. The system prioritizes quality over volume: each email is personalized with real data from the prospect business. The 4-step sequence maximizes response opportunities while respecting domain reputation with progressive warm-up.
The same, for your business
- A clinic that shows up first when people search «dentist near me».
- A garage that gets calls from Maps.
- A law firm that comes up when someone asks an AI assistant.
Technical details
Architecture
- Lead enrichment from 6 sources: Google Maps, reviews, PageSpeed, SEO analysis, SSL, and competitors
- Evaluator-optimizer pipeline: generates email, evaluates quality with Chain-of-Thought, regenerates if below threshold
- Progressive warm-up from 10→20→30 emails/day to protect domain reputation
- PostgreSQL-based processing queue: prioritizes leads with the most enriched data
- GMass webhooks for real-time tracking: opens, clicks, replies feed into analytics
Technical decisions
- Claude Sonnet 4.6 for email generation: superior quality justified by direct conversion impact
- Evaluator-optimizer pipeline with Chain-of-Thought: quality scoring before sending
- Progressive warm-up 10→20→30: protects long-term deliverability
- Apify for Google Maps: real review data without direct scraping
- LSSI Art. 21 compliance: documented legitimate interest by sector
How it was built
Cold prospecting system built on Supabase Edge Functions with pg_cron orchestration. The pipeline researches each lead by querying Google Maps (real reviews via Apify), analyzes their website (PageSpeed, SEO, SSL) and generates personalized emails using Claude Sonnet 4.6 with an evaluator-optimizer pipeline. The 4-step sequence (introduction, value, use case, last chance) executes with 3-5 day spacing. GMass handles delivery with progressive warm-up and tracking webhooks.
Tech stack
- Claude Sonnet 4.6
- Supabase Edge Functions
- GMass
- Apify
- PostgreSQL
- pg_cron
Features
- Deep research on every prospect - reviews, PageSpeed, SEO, SSL, competitors
- Emails that feel handwritten - personalized with real business data
- 4-step sequence with automatic follow-up until you get a response
- Progressive warm-up: 10→20→30 emails/day to protect domain reputation
- LSSI Article 21 compliance - legal and responsible prospecting
- Full webhook tracking: opens, clicks, and replies in real time