Automated Listing Tracker & Analysis Bot

The Problem

Manually searching second-hand listing sites every day and checking damage records/replaced parts one by one was time-consuming, and the sites aggressively blocked bot traffic.

The Approach

Built browser automation with nodriver and stealth plugins that mimics human behavior (randomized waits, back-button navigation). Each listing is parsed by an LLM into structured JSON (mileage, year, damage, replaced parts) and matched against user criteria. Moved all configuration into Telegram bot commands.

Engineering Highlights

• Stealth browser automation defeating bot detection with human-like navigation patterns.
• LLM-based extraction of structured damage/expertise data from free text.
• A full 13-command Telegram bot interface — add new radars, update criteria, stop/resume.
• SQLite-backed at-least-once deduplication plus per-run observability (run log).