I designed, deployed and maintain

I designed, deployed and maintain two self-hosted AI-agent automations on my own VPS (n8n + PM2 + nginx + SSL), both built on the same reusable pattern: source targeted contacts, personalize outreach with AI, send it, and never contact the same person twice.

Automation 1 — Cold outreach for a SaaS product:

Pulls fresh leads from the Apollo.io API (filtered by role, company size, industry), with pagination so it never re-fetches the same pool

Reveals verified emails via Apollo's bulk enrichment API, batched to respect rate limits

Generates a personalized email per lead with the OpenAI API, referencing each company's actual context

Sends from a dedicated subdomain (SPF/DKIM/DMARC configured) to protect the main domain's reputation

Automation 2 — Targeted outreach with a document attachment:

Same sourcing/personalization pattern, extended to attach a file per recipient (resume/proposal) and send via Gmail/SMTP

Runs on a fixed weekly schedule without manual intervention

Shared infrastructure across both:

Persistent, file-based deduplication so no contact is ever emailed twice, even across scheduled runs

Rate-limited, chunked API calls to stay within provider limits

Full VPS deployment and DNS/SPF/DKIM setup, done myself

Debugged real production issues along the way: rate limits, JSON-templating edge cases in n8n, and an index-alignment bug between async AI output and source data — the kind of reliability problems that actually show up once an "AI agent" runs unattended.