We’re excited to reveal that Unruly’s data scientist Alberto Boschetti will be speaking at the London Data Science Festival on April 25.
The Data Science Festival is a week of free to attend Data Talks across London from leading companies including Facebook, The BBC, Revolut, Publicis and Royal Mail. It gives over 2,000 data enthusiasts a chance to learn, grow, network and connect with the DSF community.
Alberto’s talk ‘Serving billions of real-time dynamically scaling predictions in Python: a DevOps and Data Science venture’ will take place on the festival’s main stage with timings yet to be announced.
At Unruly, we serve thousands of predictions per second with an HTTP API, with a maximum latency of less than 20ms, using an infrastructure we developed called Model Serving Layer (MSL), built principally in Python.
The MSL is able to host all the machine learning libraries, including Sklearn, Tensorflow, PyTorch et similar and has been built to (but not limited to) deliver predictions at scale for the advertising industry. It also logs traffic and monitors performance in real-time, and is able to scale up and down according to the number of requests.
The talk will explore the various design choices Alberto’s team faced when building this infrastructure, the limitations introduced by Python, and the solution they came up with.
Alberto Boschetti is a data scientist with strong expertise in signal processing and statistics. He holds a PhD in Telecommunication Engineering and is one of the founders of the data science department at Unruly.