Supplier Catalogue Automation Pipeline by Prateek SahniSupplier Catalogue Automation Pipeline by Prateek Sahni

Supplier Catalogue Automation Pipeline

Portfolio project
Prateek Sahni

The Problem

Listing one supplier product means finding it in a 373-page price list, typing a price for every size and fabric, chasing swatch images the supplier never publishes, and cropping photos by hand. One bed is 2,340 price rows, so a single catalogue update runs into days and a wrong cell is invisible.

What I Built

A Python pipeline that reads the supplier's own files and produces everything the website needs.
PDF extraction: A 373-page price list is read from the position of each word on the page, so the 70 prices printed for one bed become 2,340 rows across 5 sizes and 356 colours, taken from the supplier's public finishes catalogue. A modular sofa is priced from its spreadsheet by adding up the modules in each layout, and mirrored layouts are checked to cost the same.
Image processing: Photos are taken from the supplier's site at full resolution and re-saved with the supplier's own JPEG settings, so zooming shows the same detail as the original. Each image is measured against its source (56–65 dB, where 40 dB is visually lossless). Where a supplier publishes no swatches, the tool reuses the images the retailer already hosts, matched on colour code.
AI copy: Descriptions are written by NVIDIA Nemotron from the supplier's own text and checked before use. Prices never come from the model.
Drive workspace: Price lists live in a folder per supplier and products are rows in a sheet. A new catalogue is picked up by its file name, each run compares every price with the last and writes only the rows that changed. A React page on FastAPI lets anyone ask the assistant a question or start a re-run.
Pipeline run
Pipeline run
Swatch images
Swatch images
Independent audit
Independent audit

The Outcome

Three products run end to end: 2,340, 768 and 560 price rows, 16 photos and 632 swatch images, all six import sheets filled. Every price was verified cell by cell against the supplier documents by an independent audit pass. 197 automated tests, 89% coverage, CI on every push.

0 likes · 0 views · Nov 15, 2026

Python pipeline that turns supplier price lists, swatch cards and websites into import-ready e-commerce data. 197 tests, independently audited.

Timeline

Jan 1, 2026 - Ongoing

Tags