SonarFetch
SaaS platform that unifies data from 400+ manufacturers. Automates what took months of manual work in minutes.
Context
Steliau Technology is a French electronics distributor — resistors, capacitors, microcontrollers, sensors — working with over 400 different manufacturers. To keep its catalog current, their teams had to manually open each manufacturer's website, copy product references, technical specs, prices, and re-enter everything into their system. A fully manual process, repeated for every manufacturer, with every product update. SonarFetch grew out of this observation, and from the work I started during my internship at Steliau as an initial PoC.
Problem
In practice, here's what it looks like: an employee opens a manufacturer's site, searches products one by one, copies data into a spreadsheet or directly into the back-office. Times 400 manufacturers. Times regular updates when references change, products are discontinued, new ones arrive. It's industrial copy-paste — time-consuming, error-prone, and impossible to keep current. Traditional scraping doesn't solve it either: you need a hand-written script per manufacturer, which breaks as soon as the site changes its layout. With 400 manufacturers, that's 400 scripts to maintain.
Approach
The core idea: a scraper that isn't hard-coded but understands and adapts. Instead of one script per manufacturer, a single agent that analyzes any site's structure and knows what to extract — no manual configuration needed. The agent receives a manufacturer website URL, explores the navigation, identifies product pages, extracts the data, and normalizes it to a unified schema. All manufacturers finally speak the same language: same field names, same units, same structure — regardless of the source format. SonarFetch is the direct evolution of the PoC started at Steliau, which relied on classical ML (NER, neural classification). In SonarFetch, that ML gives way to a fully agentic LLM architecture: more robust, more flexible, and no retraining needed for each new manufacturer.
Impact
Adding 200 new products from a manufacturer? What took 3 weeks of extraction and cleanup now takes 15 minutes of human validation. Data is normalized across manufacturers — a customer searching for a 5V component can compare products from different vendors without getting lost in proprietary naming conventions. And because everything is consistently structured, SonarFetch becomes the foundation for high-value features: automated pricing comparison, direct ERP connection, storefront upload, competitive analysis.