Search must understand the buyer, not only the database.
Auto Delovi RM in Zaječar sells body parts throughout Serbia. A buyer knows a make, model and year, while the system has to connect that description to the right part and preserve the choice through the journey.
Tail lights · vehicle choice preserved
A catalogue is a tree, not a flat list
Over 1,800 parts connect to 26 categories, 69 makes and about 1,794 models. Permanent pages exist for categories, makes, models and their intersections, so search and sharing do not depend on a temporary filter.
The query is split into intent
Empty words are ignored, a year becomes a filter and a part number matches with spaces or hyphens. Narrow synonym groups connect common Serbian terms without returning half the catalogue.
- years and ranges
- part numbers
- Latin, Cyrillic and accented letters
Ranking preserves meaning
A match at the start of a name ranks above a secondary occurrence. A query for a front bumper therefore appears before a reflector located in a rear bumper. When nothing matches, the catalogue does not pretend by showing every item.
Vehicle choice remains in the address
Make, model, year and query enter the URL through browser history. Sharing, returning from a product and opening from Viber or Facebook restores the same context instead of an empty catalogue.
Scope and evidence boundary
The system uses Next.js 15, Node.js and SQLite, with a panel for the owner to manage the offer. The case describes published scope and behaviour without claiming an unverified sales outcome.