Cold Email Outreach Automation
Scraping pipeline with LLaMA-personalized emails and automated Gmail delivery.
Overview
Built an end-to-end cold email outreach automation that scrapes company data from target industries, generates personalized email copy using LLaMA, and sends them through Gmail with automated follow-ups.
Business Problem
Manual cold outreach is time-consuming and doesn't scale. Generic templates get low response rates. Personalized outreach requires research on each prospect, which is prohibitively slow at scale.
Solution
Created a pipeline that scrapes company information from public sources, enriches each prospect profile, generates personalized email copy using LLaMA 3.3 70B with company-specific context, and sends through Gmail with a 3-email follow-up sequence.
Architecture
Key Features
- →Web scraping pipeline for company profile enrichment (industry, size, tech stack, recent news)
- →LLaMA-generated personalized email copy incorporating company-specific context
- →3-email follow-up sequence with configurable delays
- →Gmail API integration with send quota management
- →Open/click tracking via pixel tracking
- →Response detection and auto-stop on reply
Gallery
Complete pipeline architecture showing scraping, personalization, and email delivery components.
Follow up email workflow with n8n and Gmail API integration
Tech Stack
Engineering Challenges
Email Personalization at Scale
LLM-generated emails risk being too generic or hallucinating company details. Built a prompt architecture that injects verified scraped data points and constrains the model to only use provided context, never inventing details.
Gmail Send Quotas
Google's daily send limits require careful scheduling. Implemented a rate-limiting layer in n8n that spaces sends across the day and respects API quota limits.
Response Detection
Automatically detecting human replies to stop follow-ups requires parsing reply semantics, not just subject matching. Used a lightweight classifier to distinguish auto-replies and out-of-office from genuine human responses.
Lessons Learned
- →Highly personalized emails (company name, recent news, specific pain point references) outperform generic templates by 3-5x on reply rates
- →A 3-email sequence with 2-day intervals maximizes response rate without being intrusive
- →Auto-stop on human reply is critical for maintaining sender reputation
Results
Achieved 32% reply rate and 8% conversion rate on a 500-prospect campaign, with the system autonomously managing the entire outreach lifecycle.
Related Projects
Let's Build Something Similar
Have a project in mind? Let's discuss how we can architect and build it together.
Discuss Your Project