An automated lead generation engine that uses AI vision models to analyse hundreds of websites simultaneously, flagging ones that look visually outdated. For each flagged site, the engine finds a contact and sends a tailored cold email. Runs autonomously 24/7 — continuously filling a pipeline without manual work.
Most lead generation is manual — searching, qualifying, writing emails. This system automates the entire chain. Puppeteer visits hundreds of sites per session, takes screenshots, and sends them to a vision model that scores the design quality from 1-10. Sites scoring 8+ are flagged as targets.
For each flagged site, the engine finds the contact email, generates a personalised cold email based on the specific issues detected, and queues it for sending. The whole pipeline runs unattended.
Multiple Puppeteer workers run in parallel — each visiting, screenshotting, and scoring sites independently. Results feed into a central queue. The email composer reads the AI analysis to craft personalised messages referencing the specific issues found on each site.
Getting consistent, structured scoring from an AI vision model across hundreds of visually diverse websites required careful prompt engineering and output validation.
Running hundreds of Puppeteer instances simultaneously without memory exhaustion or IP bans required worker pool management and rotation logic.
Finding a real contact email for each flagged site — avoiding generic info@ addresses where possible — using a combination of DOM parsing and heuristics.
Every email needed to reference something specific about the site — not a generic template — to achieve meaningful reply rates. The AI analysis output drives this directly.