Rino Montiel has been working in the tech industry for over 10 years building systems and data pipelines. He’s passionate about automation, developing high performance systems and teams, and using technology to build products that assist humans with day-to-day tasks.
Today, Rino leads the Platform and the Data teams at EyeEm – building systems that run e-commerce, photo sharing, search and image classification, in addition to creating data pipelines for analytics and machine learnin
November 11th
12:00
Talk
English
Presenting the infrastructure work that we did to support the launch of The Roll, our smart camera roll app.
We’ll give a brief overview of deep learning at EyeEm, and how we use it to build our products. This includes how we use it for both search, image captioning and aesthetic ranking.
We needed to build a system, exposed through an API and tracked through a data pipeline that could run our deep learning models on uploaded images, return a variety of data to the users (keywords detected in the photo, photo scores) and track all the data for later analysis.
We’ll talk about how we optimized for throughput by load testing memory pressure on the GPUs we used on AWS instances, and how we go from training a model to putting it in production.
On launch day, we were expecting to have to process around 500 photos/sec. In reality, we peaked at around 850 photos/sec, without any impact on performance.
