← All work
bazos-agent
Watches 248,000 car listings and ranks the ones worth a phone call
- Status
- ToolInternal tool, runs hourly
- Period
- 2026–now
- My role
- Sole developer
- Stack
- Python · asyncio · PostgreSQL
The problem
Sourcing stock for a dealership means reading classifieds all day. Bazos.sk carries around 248,000 active car listings across 41 brands, and the good ones are gone in hours.
How it was built
Backfill once, then only look at what is new
A full scrape builds the baseline; after that the agent runs in delta mode and finishes a cycle in under three minutes. Rescraping everything hourly would be both slower and rude.
A composite score, because one signal lies
Price alone finds salvage. The ranking combines price against the model's own distribution with mileage, year, and listing quality, so what surfaces is a deal rather than a mistake.
Where it landed
- Full cycle over new listings in under three minutes
- Feeds Fleetron as its external-listings sourcing module, and runs standalone too
Scraping is the easy half
Getting the pages is a solved problem. Deciding which of 248,000 cars is worth a phone call is the part that needed the work.