"We have a perceptual AI, now what?" At Arc XP, we faced a daunting challenge with our video infrastructure: we trained a machine learning model to pick a thumbnail for a video, but due to the nature of its construction, deploying the model at scale proved difficult. Using the model against one or two videos a day was no problem, but what about thousands?
Orchestrating the machinery that makes the AI work at scale: The entire perceptual pipeline has many moving parts, including mechanisms for scene detection, key frame extraction, AWS resource manipulation, and, of course, the machine learning model itself. We used Temporal to bring order and sanity to this process, boiling the entire process down to a single, digestible, durable, and repeatable workflow.
Working closely with the Temporal team themselves (shoutout to Chad Retz!), we identified areas for improvement and “correctness” in what should (and shouldn’t) be an activity and how “durable” each part should be.
My name is Eric Ziegler, I am a principal software engineer at the Washington Post, specifically working in the Arc XP division on our Video Center product. My favorite programming languages are Rust, Python, and Java. As far as favorite technical topics, I love machine learning, video engineering, and production durability (who doesn’t like a silent pager?). When I’m not programming, my wife and I are raising foster kittens, and I dabble in a bit of woodworking.
Ready to learn why companies like Netflix, Doordash, and Stripe trust Temporal as their secure and scalable way to build and innovate?
Financial Services
Go
How ANZ Bank accelerated home loan origination with Temporal