About Us

Decades of Experience as High Tech Strategy and Development Consultants

Gradient Valley was co-founded by Keith Bourne and Dave Koziol. Keith and Dave have been working together for almost a decade, both bringing multiple decades of experience as high-tech consultants, strategists, and developers. We’ve helped hundreds of companies to successfully implement their technology strategies in web, mobile, and machine learning. Dave is the founder and president of Arbormoon, which has focused on web and mobile projects for over 20 years. Gradient Valley will utilize many of the resources Arbormoon has to offer, including a large network of developers with every skill set we could possibly need. This gives us the flexibility to handle any size project in any time frame that is required of us.  

Perfect Combination of Business Strategy, Machine Learning, and Practical Experience

A machine learning project requires a much deeper understanding of the underlying problems that are trying to be solved, compared to your typical software development project.  You can't just throw a spec at some developers and data scientists and expect they will come up with the optimal solutions for your business oriented problems.  Disconnect between business strategy and the development team is often cited as the biggest challenge for machine learning projects.  When this happens, it is typically because the team working on the machine learning project has a difficult time grasping the business context of the problem they are trying to solve. Gradient Valley brings together business strategists and machine learning so that we can fully understand the business problems your organization is facing. We then apply machine learning with this understanding, coming up with a solution that has a much higher chance of success.


It is also quite different to develop a machine learning project in isolation, as is typically done in the academic work, from developing a model that works and continually improves itself in a production environment with live streaming data. This is yet another reason many machine learning projects fail. This is an unfortunate and very expensive mistake, and quite often the ML development team doesn't understand why it happens because it worked fine for them back at school. So why does this happen? In the "real" world, data and the models developed from that data, are not static. They are constantly flowing into an organization, changing, adjusting, responding to all kinds of different external forces. You must start from day one handling data in the same way it will be handled in production (streaming), and developing your model in a way that accounts for that. This takes an entirely different approach from an infrastructure standpoint, and is much more expensive to change after you realize this compared to starting with it in the first place. Gradient Valley understands this, and so we start our projects from day one with the understanding that our models need to constantly evolve based on new data. And ultimately, this works best for the client, because the entire project is set up in a way that is easy to hand off at any time. 

Cost Effective Approach

Machine Learning projects do not have to be nearly as expensive as they often are.  And spending more on a project does not guarantee success. In fact, it often is a red flag that you are doing it wrong.  Gradient Valley has learned how to significantly reduce the costs of the typical machine learning project by utilizing the latest techniques, infrastructures, and project management approaches, without sacrificing any quality of service.

What is a "gradient valley"?

Gradient Descent is a method often used in data science (as well as math in general), to optimize the model that you ultimately build for analyzing new data. To sum it up, it finds the global minimum in our function, where the lowest value is the optimal value for us.  As shown in this graphical representation, there are peaks and valleys in this key value.  The valleys are the optimal points, so we seek to optimize the model using various forms of gradient descent to find the best of those valleys.  So, as we say at Gradient Valley, it's nice in the Valley!  The Gradient Valley.