Overview

What it is: An automated lead generation + deep personalization system that finds leads, verifies + enriches contact data, researches each company website, and drafts hyper-personalized cold emails.

Who it’s for: B2B outbound campaigns targeting niches like Solar/Renewables, YC startups, and Home Services.

Core idea: Use AI for high-leverage reasoning and writing, but keep execution deterministic and reliable via scripts.

Architecture (3-layer system)

• Layer 1: Directives (directives/)
  • Markdown “SOPs” that define campaign rules, tools, and quality checks.
• Layer 2: Orchestration (AI agent)
  • The agent reads directives and delegates work to scripts.
  • Acts like a manager, not a monolithic do-everything model.
• Layer 3: Execution (execution/)
  • Python scripts for scraping, API calls, verification, and data ops.

How it works (pipeline)

1. Scraping & sourcing
   • Uses Apify to scrape sources like LinkedIn, ProductHunt, and YC directories.
   • Example scripts: run_apify_actor.py, scrape_yc.py, scrape_producthunt_90d.py.
   • Lead targeting is driven by job titles (Founder/CEO) and keyword filters.
2. Verification & filtering
   • Removes bad contacts early using concurrent verification.
   • Checks include syntax validation and MX record lookups.
   • Example scripts: verify_emails_concurrent.py, verify_leads.py.
3. Enrichment (finding missing emails)
   • For incomplete leads, runs enrichment with external APIs (ex: Anymail Finder).
   • Example scripts: batch_enrichment.py, enrich_startups_anymail.py.
4. Deep research & personalization (the “secret sauce”)
   • execution/deep_research_main.py aggressively scrapes each company’s website.
   • Uses GPT-4o to:
     • Understand positioning and business model.
     • Identify a hidden pain / inefficiency.
     • Generate a “sniper” subject line + hacker-toned personalized email.
   • Extra social dynamics: add_cofounder_snippet.py can add a cofounder icebreaker.
5. A/B testing (simulation)
   • Runs large-scale simulations against AI personas to predict reply-rate lift.
   • Example: ab_test_personalization.py.
6. UI / Operations
   • CLI run: python execution/master_run.py --limit 2000.
   • Local web UI:
     • app.py runs Flask on http://localhost:5000.
     • Trigger runs and watch live output in-browser.

Key files

• README.md + CLAUDE.md: project philosophy + setup
• app.py: Flask UI
• execution/master_run.py: end-to-end run (scrape → verify → enrich)
• execution/deep_research_main.py: AI personalization engine