Demo build · AI Automation

AI Lead Generation and Outreach Autopilot

A scheduled pipeline that finds local businesses city by city, writes them to Google Sheets and sends each one a personalised cold email, written by a language model and sent only during the recipient's business hours.

The problem

Outbound sales to local businesses is slow manual work: someone searches maps, copies details into a spreadsheet, writes an email per lead and keeps track of who replied. It rarely gets done consistently, and generic templates get ignored.

How it works

Area planning. For each city in a queue, a language model proposes the neighbourhoods worth searching and returns them as structured JSON.
Lead collection. An open-source Google Maps scraper runs locally in Docker. Results are cleaned into lead fields (name, category, address, phone, website, rating) and enriched with social links from each business's site.
Storage and reporting. Leads go into Google Sheets, one tab per country, with a dashboard, sent-email log and activity tabs refreshed on every run. Pipeline state lives in SQLite.
Personalised outreach. A language model writes the opener and email for each lead from its details. Sending respects the recipient's local business hours and pauses when bounces rise.
Feedback loop. Replies, unsubscribes and bounces are detected over IMAP and written back, so nobody is emailed twice or after opting out.
Runs itself. A scheduler fires the pipeline every 15 minutes, with a lock file so runs never overlap.

What it demonstrates

A language model used only at the steps that need judgement (where to search, what to say), with ordinary, testable automation everywhere else: queues, deduplication, rate limits, compliance rules and a human-readable dashboard.

Built with

Python, an LLM API, Playwright, Google Sheets API, Gmail SMTP/IMAP, SQLite, Docker and a system scheduler. The scraping engine is the open-source gosom/google-maps-scraper (MIT).