One of the challenges that we've faced while building Emprezzo is finding a way to properly search products from hundreds of different sources. Products aren't titled, tagged, or described in a standard manner. That's made it hard to define the initial search fields, and the weighting of the fields is even more difficult. While one retailer may describe a products as "Women's Purple Mountain Ski Socks", a company that only makes ski socks make name the product "Purple Mountains".

To improve the results, we're adding some machine-learning in by using computer vision to tag the results. There are tags on roughly 20% of the 40,000 product catalog now and the results have been quite accurate in our testing thus far. And while the accuracy is important, it's perhaps more important to add a standardized process in to review the products from the same perspective.

The product above wagged with the keyword Clothing, T-shirt, Sleeve, Photo shoot, and Yellow. While the tags are not perfect, they are a big improvement when applied in an equal manner across all products.

It's amazing how simple its become to access and deploy technologies like machine-learning based computer vision. We're in the process of updating 40,000 products from hundreds of different retailes and the results will be standardized in a way that would have been completely infeasible in a manual manner. The results will serve as an additional factor in improving the overall search experiences at Emprezzo.