How Babylon Health is building an enterprise-ready analytics platform | Data





Babylon Health is one of the fastest growing healthtech startups in the UK, giving users quick access to NHS GPs via video calls within its app, as well as an interactive symptom checker, which is underpinned by diagnostics data and a chatbot interface.

Founded by Ali Parsa in 2013, and having raised more than $80 million in funding to date, Babylon and rival apps like Push Doctor are quickly changing the way UK consumers access health advice and has even been talked up by the health secretary Matt Hancock, a move which drew heavy criticism from the healthcare industry.

© Babylon Health
© Babylon Health

Services like the symptom checker and Healthcheck – another Babylon service which builds a health report for users after asking them a set of personal questions – are essentially underpinned by a massive medical database with a set of algorithms built on top to give personalised recommendations.

This gives the company a whole range of analytical use cases to meet, from standard operational business intelligence (BI), to more complex streaming analytics and machine learning or data science-led cases.

To support this side of the business, the startup hired Neil Winters as director of business intelligence in October 2017. Winters previously held roles as BI project manager at BBC Worldwide and then as a senior BI project manager at Microsoft and the EMEA BI manager at Apple.

Why did he decide to join a startup like Babylon at this stage in his career?

“It was very much the opportunity to drive something from the beginning,” he told Computerworld UK last week during the Dataworks summit in Barcelona. “The great thing about Babylon, working there is we are still a very young company, there’s not a huge amount of legacy processes and just the culture of a startup around build fast, be brilliant and innovative.”

Building a scalable analytics platform

Winters was essentially tasked with building a scalable analytics platform that could grow securely with the business.

“We have ambitions to be a very large company,” he said. “That means we have to be an enterprise-thinking company, as well as an innovative company. Unfortunately enterprise thinking comes with very boring things like security, regulations, auditability, control, and maintenance. Some of those elements are missing from other pure open source solutions.”

Babylon opted to partner with Hortonworks, a vendor specialising in enterprise versions of open source big data technologies, like Hadoop, which recently completed its merger with its old rival Cloudera.

Read next: Everything you need to know about the new Cloudera Data Platform: vision, migration and roadmap

“The nice bit about the Hortonworks and Cloudera stacks now is that it becomes integrated. I build one layer, one ingestion flow, I can tag it, flag it, audit it and reuse it quickly across use cases and control it with a single lens,” Winters said.

This is all run in the cloud with AWS and is bolstered by anonymisation services to ensure no personally identifiable data is present in the analytics, with encryption at every level.

“As we become an enterprise company working in health, we are going to get a lot more focus, rightly so, from all different sort of regulators and organisations across the world, so it’s very much it was a mixture of: yes, we need all the great, cool capabilities, but we also need to continue thinking about what’s our enterprise footprint going to have to look like,” he added.

And is he worried about the impact the merger will have on his decision to invest in Hortonworks technology?

“It probably will be a bit of a headache,” he admitted. “I think on both sides they’ve got strengths and weaknesses on their tools and bringing them together to choose the best one was probably an interesting fight between the two organisations.

“I think it’s a fact of life using any technology and specifically open source technologies that some will live in and some will die, you have to look at continuously what the future is holding. I think we’re still okay, there’s a lot of flexibility in the organisation. I would be more concerned if I was a massive bank and had spent the last 10 years building out something huge.”

Building a platform for AI

As a startup, Babylon doesn’t face the sort of petabyte-scale challenges many enterprise customers come to Cloudera to solve. Rather, “some of the biggest challenges we have is actually complexity, so the variety and velocity of data, rather than volume,” Winters said.

“We have a lot of different data coming in across lots and lots of different areas. It can be lots of textual data, with streams of different information coming in, there’s links to our diagnosis engines behind the scenes, which, again, are very complicated in what they’re doing.”

Going into a bit more detail, Winters defined how Babylon has built an analytics platform that can serve these needs, moving to what he calls a more “service-oriented architecture”.

This involves all analytics being pushed as an event into a central Kafka queue

He added: “We very much have a practice which we’re developing within Babylon, which is as new features, new functions, and everything new is added, that should be the default position: what does the event look like? What does the information mean? How’s that going to be used and consumed for the different analytical use cases going further downstream? So we’re very much moving towards that, event model and defining data as the assets.”

When it comes to working with advanced users, the sort of data scientists developing deep learning models that will produce diagnostic advice, this process “becomes a partnership as much as possible”.

“We on the engineering side of things, on the platform, should do as much heavy lifting as we can so the data scientists don’t have to do that ETL process and joining stuff up, which isn’t a good use of their time and is expensive for them to do that work,” Winters said.

Babylon is also currently building its research capability, with Winters looking to give those people a platform where the data is effectively ring-fenced, allowing the data scientists to freely experiment.

“They can build whatever they fancy using some aggregated data from external places, then take that data and run it against the real data on our platform and see if the theoretical model works when run against real data,” said Winters.


Leave a Reply

This website uses cookies. By continuing to use this site, you accept our use of cookies.