"Built an autonomous pipeline that cold pitches local businesses and delivers paid diagnostic reports while I sleep

I've been building in public for a while but this one feels different. I want to share what I actually built, how it works, and what's happened in the first 12 hours.

The problem I was solving

Local service businesses (HVAC, law firms, roofing, veterinary, etc.) are hemorrhaging customers online and most of them have no idea. Bad mobile speed, no after-hours capture, weak reviews, missing schema markup. The gaps are observable from public data. I wanted to productize the diagnosis.

What I built

A fully autonomous pipeline called Apex Logic v9 that:

1. Hunts leads using Google Places API — targets local service businesses by industry and city
2. Scores them across 4 pillars: Reputation, Lead Capture, After-Hours Coverage, Website Health
3. Runs a full LLM audit using a local Ollama model (qwen2.5:32b on my RTX 3090) — grades each business A-F and estimates annual revenue leakage
4. Generates a personalized cold pitch — the email references their specific grade, their specific top finding, and their specific dollar loss estimate. No generic templates.
5. Sends the email via IONOS SMTP with jitter and account rotation
6. Tracks opens via a 1x1 pixel and triggers a follow-up sequence automatically when someone reads the email
7. Handles replies — an IMAP monitor reads incoming replies, classifies intent with Ollama (interested / question / objection / unsubscribe / consult), and drafts a contextual response using the lead's diagnostic data
8. Takes payments via Stripe — $149 snapshot or $399 full report, collected through a checkout modal that passes business metadata so the pipeline knows who to run for
9. Delivers the PDF automatically after payment — no manual steps

The whole thing runs in 3 screen sessions on a WSL2 Ubuntu machine at my office.

First 12 hours

• 49 cold emails sent
• 157 leads in the pipeline with $870k–$2.4M in identified revenue leakage across active prospects
• 1 confirmed open — Central Houston Animal Hospital
• Follow-up automatically sent 4 hours after the open with subject line "Re: Central Houston Animal Hospital scored a C — no after-hours lead capture"
• 0 replies yet but it's been 12 hours

The tech stack

• Python / Flask
• SQLite + SQLAlchemy
• Ollama (local LLM — qwen2.5:32b)
• Google Places API
• IONOS SMTP / IMAP
• Stripe Checkout
• ReportLab (PDF generation)
• ngrok (webhook tunnel)
• Screen sessions (yes, really)

What I'm proud of

The personalization. Every email leads with the business name, their actual grade, their actual top finding, and their actual estimated dollar loss. The subject line for Central Houston Animal Hospital was "Central Houston Animal Hospital scored a C — no after-hours lead capture." That's not a template. That's their data.

What's next

• Scale daily send volume with additional email accounts
• First sale validation
• Building toward a fully hands-off revenue engine

Happy to answer questions about the architecture, the LLM prompting, or the cold email approach.