Sep 2026 – Oct 2026
London Spot product pipeline
Turns POS stock and Amazon liquidation manifests into complete product cards in Georgian and English
- AI
- AI data pipeline
- Duration
- 1 month
- Company
- London Spot
- Role
- Data and automation engineer
Overview
A resumable pipeline that turns the shop’s POS stock and Amazon liquidation manifests into complete product cards in Georgian and English, with names, descriptions, photos, categories and prices. It loads them into the catalogue, so listing a new shipment no longer means translating and sorting by hand.
Tools
- Node.js
- Gemini
- Bright Data
- PostgreSQL
- Docker
The need
Every shipment brings a large lot of Amazon liquidation items, mostly single units. Listing them meant translating Amazon descriptions, collecting photos and placing every item in a category by hand, and the items on the shop shelf had to be identified in the POS one by one.
The steps
The Node.js pipeline pulls the shop’s stock from the POS or parses a 45-column manifest, repairing broken encoding. It matches items to Amazon through Bright Data search with Gemini checking each match, fetches products from several Amazon marketplaces with fallbacks and merges their categories into one tree based on amazon.de. Gemini then translates everything into Georgian and English and writes search-friendly Georgian names; finally the units are priced, merged, loaded into PostgreSQL and exported to the site or imported into the live admin.
Safe to rerun
Every step is cache-first and resumable, so a run can stop and continue without paying twice for the same data. Paid runs need an explicit spending cap, product ids are frozen before any renaming, and the steps are covered by tests.
Result
5,742 units (4,886 products) in 22 departments, 176 subcategories and 2,395 brands, all translated with 0 failures. The Georgian translation cost about $4 in Gemini calls, and a naming run rewrote 4,268 names and 3,519 descriptions for $3.30 with ids and prices unchanged.