How Exscientia reduced the time it takes to monitor and prepare models from days to hours.

About Exscientia

Exscientia plc (EXAI (NASDAQ)) is an AI-powered drug discovery organization whose mission is to design and develop novel, precision-engineered drugs with an improved probability of clinical success for the benefit of patients. AI is both at the heart of their business and is the determining factor in the success or failure of projects. This means the accuracy and stability of models is a high priority.

What was the challenge?

Model deployment at Exscientia is unique because there is no human interaction in the training and deployment of models. They are created autonomously from data pools. The entire deployment process is automated resulting in thousands of models being delivered, monitored and retrained.

When Sash Stasyk, MLOps Engineering Team Lead, joined the team in Summer 2021, models were being deployed using an open source deployment solution and then monitored with another open source inference platform both of which were suitable to their requirements at the time. As Exscientia has grown and expanded its reach and goals, so has its need for enterprise-grade scale. The team was looking for additional operational efficiencies and other ways to debug and stabilize models.

How did they do it?

Sash and his team began looking for additional ways to continue to advance the technology, including ways to advance how the team was looking at cases where outcomes don’t match predictions. He had already heard of Seldon and began the evaluation process just as Adrian Rossall joined the team. Adrian had also used Seldon to great success before and made it a priority to start the implementation quickly.

Sash chose Seldon Deploy, to assist with monitoring and managing models in production in order to enable them to scale. Seldon would speed up the debugging process to uphold the stability and reliability of those many models now in production.

“I liked that it was modular and we now have the observability in place and were able to simply port all our models across. It was painless to get started.”

Exscientia had to make a choice whether to change their model code or not when moving to a new MLOps platform. This was a pivotal decision as changing the code could have a detrimental slow down effect on the entire project. They decided to keep the same code and port over all their existing models. One benefit of implementing Seldon was that the code didn’t matter to Seldon and they were able to fit the architecture to the model code. That theory paid off and saved a lot of time.

What's next for Sash and Exscientia

With Seldon Deploy now in play Exscientia have reduced the time to debug and resolve a trivial issue from hours to minutes and a serious issue anywhere from 24 hours up to just an hour on average. The impact doesn’t just stop at the Data Science and Machine Learning teams. Drug chemists can now make decisions in the design cycle much faster resulting in oncology drugs getting to market quicker saving lives of cancer patients. 

"Seldon has made huge difference to how we scale and deploy our inference Ecosystem."
Sash Stasyk
MLOps Engineering Team Lead

Download the PDF Version

Add your email here and we will send a copy of the PDF right to your inbox.

See it for yourself

Serve, monitor, explain, and manage your models today.

© 2023 Seldon Technologies. All Rights Reserved.