Building a better search engine for Second-Hand Markets

Goal:

The goal of this project was to create a better search engine for second-hand markets that enabled comparing (new x used products) prices in a single place

Impact:

By building an index and search engine that could identify products in second-hand markets and compare prices on a single page. The project was able to process hundreds of thousands of ads per week and have a product catalog of more than 1,000 electronic items. This made it easy for users to compare "used x new" product prices, saving them time and money.

Problems:

The project faced several challenges, including poor search features on second-hand markets, the sheer volume of ads in Brazil, and the highly unstructured nature of the data. Additionally, there was an issue with product quality, as used products may be cheaper than new ones but be broken (or have parts missing).

Solution:

To overcome these challenges, the project used natural language processing (NLP) to process the ads and match them to the product catalog. A deep learning pricing model was created that leveraged the NLP understanding of the ad description to price each ad and rank them in terms of bad/good deals.

Technical Highlight:

One of the key technical highlights of this project was the use of NLP and deep learning to process and understand unstructured data from second-hand market ads. This allowed for accurate matching of products to the catalog and pricing of each ad, despite the challenges posed by the data. Additionally, the development of a pricing model that could take into account the quality of the used products was a key achievement of this project.