Getting Machines to Think Less Hard
Building robust data pipelines for heterogeneous or inconsistent data sets is notoriously difficult, time-consuming, and usually unscalable. Seemingly simple features such as "patient age" or "a hospital admission event" are quickly exposed to be frustratingly complicated. Especially when the data problem is solved or ignored in subtly different ways. The problem is multiplied when devising more sophisticated meta-features, groupers, or automatically generated clusters.
We have devised a node-based approach, where machines treat data work and features as abstractions or "nodes". Nodes can then be referenced, cached, logged, used, and tested in a machine learning pipeline; cleanly separating model-building from data-engineering. Powerfully, nodes can be made dependent on other nodes, allowing for layers of abstraction.
Teddy is the CEO and cofounder of pulseData. He's been a career entrepreneur, leading software product teams at fast-growing technology companies. Teddy left Wall Street in 1999 to co-found an instant messaging/co-browsing company, and has been building enterprise startups ever since (a detour through business school and Bain consulting). But none as important as this one. The pulseData team is committed to the idea that if the amazing talent in Technology and Healthcare decide to work together, there is enough data and data science to eliminate preventable sickness. We're a team of quant hedge fund machine learning healthcare enterprise product folks, trying to stop using buzz words. Oh, and using A.I.