Alpha Vertex is a year-old New York City startup with an ambitious agenda: It wants to create a graph database of global financial knowledge.
CTO Michael Bishop says the goal is to use predictive modeling to help companies judge risk and investors get insight on what drives the market. To do so has required the company to build a massive technical back-end that uses some hottest emerging technologies. Two of the most important are Google’s cloud-based machine learning algorithms and IBM’s OpenWhisk, a serverless or Function-as-a-Service platform.
+MORE AT NETWORK WORLD: Serverless explainer: The next-generation of cloud infrastructure +
Alpha Vertex’s custom-built machine takes in a lot of raw data. Any piece of information that could influence stock market movement is a potential data source – including news articles, government filings, market data, etc. At a basic level, it has created systems for IBM’s OpenWhisk platform to automatically collect data, reformat it in a way that Alpha Vertex’s machine can consume it, and send it over to Google’s cloud for analysis.
In this use case, serverless computing is an ideal solution, Bishop says. Stock market data is very unpredictable. Before the market opens and just after it closes the company gets a rush of data. Geopolitical (Brexit) or geophysical (earthquake) events create unexpected spikes.
In years past an IaaS public cloud could have been used. But Bishop says it doesn't make sense to pay for a virtual machine to be spun up 24 hours a day, seven days a week. Instead, by using a serverless computing platform Alpha Vertex is able to execute a function when an event is triggered – such as a company filing a regulatory note. The platform automatically executes the action it has been programmed to perform – for example converting a PDF into a Word file and automatically stripping out potentially relevant information, then transferring it to Google’s TensorFlow platform for analysis.
No matter how many data points are ingested into the system, IBM’s OpenWhisk executes the function it has been programmed to perform. Bishop doesn't have to spin up virtual machines, load balancers or block storage volumes. “The ability to elastically scale at increasingly granular levels of your application works well because we have a wide variety of highly specialized functions that we need to be ready to engage at any time in an unpredictable manner,” he explains. “This elasticity helps us save quite a bit of money everyday because we don’t have to over-provision. Functions just execute as they need to.”