The Problem

Many businesses have problems that machine learning can address, as well as data to drive those solutions, and yet effective ML implementations are not ubiquitous.


Accurate but not SHIPPAble

Widely available ML libraries are extremely accurate but fall short in 2 crucial ways:

  • Results are not detailed - The difference between "Normal abdomen" and "Normal abdomen, 51% of the time" can be huge. To effectively design applications, access to this information is a requirement that many frameworks do not allow or even possess. See FullRank's Probabilistic Results.

  • Results are inscrutable - In the above example, the user may want to know why the system identified a medical image as potentially abnormal: What location? What evidence is there? What is the strength of the conclusion? FullRank's results are Fully Explainable.

DoES not allow Exploration

Existing methods will create an accurate model with defined inputs but won't allow operators to guide the process dynamically.


Consider estimation of a house price - location is the most important thing, by far, but one can't change the location of an existing house; so for owners, this is not useful.


What about users who are looking for a house to purchase - they can look in many areas? Two different approaches to the same problem: how does one value a property? Users may need to explore the data and create models based on 'prospecting'.

FullRank allows this prospecting and real-time deployment of versioned models to a REST API.