Cold Email Automation

Cold Email Automation

A data pipeline + Claude-powered outreach system built to run job search at scale: extracting verified hiring contacts from a public government dataset, generating personalized cold emails with Claude, and pushing them to Gmail as drafts via a Claude agent over MCP.

What it does

• Sources contacts from public record. extract_contacts.py parses the U.S. Department of Labor's quarterly LCA (Labor Condition Application) disclosure data — a public dataset employers are required to file — and pulls the listed point-of-contact for each filing.
• Splits & dedupes. Contacts are separated into hiring managers vs. talent-acquisition/recruiting contacts, deduplicated by email, and checked against every address already contacted.
• Personalizes at scale. scripts/batch_gen.py calls the Claude API to infer a company-specific angle for each lead (based only on company name + contact title) and drafts a tailored email from a fixed template — batched, rate-limited, and resumable via progress.json.
• Pushes real Gmail drafts via Claude + MCP. Generated emails aren't sent through the raw Gmail API — a Claude agent connected to Gmail through MCP (Model Context Protocol) reads each drafted email and creates it directly as a Gmail draft.
• Follow-up sequencing (built, not deployed). scripts/follow_up_gen.py computes a 3-stage follow-up cadence (business-day offsets) per lead in follow_up_tracker.csv, ready to drive F1/F2/F3 drafts — this stage was built but not run in production for this campaign.

Scale

• 13,000+ leads processed from a public dataset of ~83,000 LCA filings
• 1,600+ personalized drafts generated
• Fully automated resume-from-last-position on every run — no re-processing, no duplicate outreach
• Landed 8+ calls with direct team members rather than screening through ATS

Stack

Python, pandas, numpy (business-day follow-up math), Anthropic Claude API (content generation), Claude Agent + Gmail MCP connector