Back
After a thorough technical audit, TechAscent was asked to architect and rebuild Fathym's weather forecasting app; meshing NOAA models with custom machine learning.
Problem
Productizing prototypes is always challenging; often the first version of a program will produce some useful results, but can be difficult to maintain, extend, and scale. With a proven business case around road condition weather modeling, Fathym was very eager to address the problems in the existing system’s ad-hoc architecture that were leading to poor performance of demonstration applications. Moreover, communication between the development team and the rest of the organization was strained, negatively impacting integration with the rest of the company.
Solution
We used our open source numerics and data science stack to build a production ready app in 4 months. The app extended functionality of the original prototype while reducing code footprint and hardware requirements.
By doing a thorough technical analysis of the existing system, augmenting documentation, and fostering communication, TechAscent built up a company-wide knowledge base around the prototype, elucidating its full business implications. We developed a partnership, with a shared toolset and domain model, to verify assumptions about the design and implementation of the system. This allowed Fathym to make open, reasoned judgments about moving the prototype system into production, realizing returns on their investment.
Following the analysis, Fathym contracted us to develop the next version of their weather forecasting system using our own open source numerics and data science stack. The next version needed to preserve correctness while drastically improving performance. With a small team, together with everything learned during the earlier analysis, we were able to build a production version in only 4 months; a fraction of the time it took to build the initial prototype. The improved system, built from first principles, scales much better due to a reduced footprint in both code size and hardware requirements. This kind of data-intensive cloud native solution (kubernetes, cloud agnostic storage, distributed computation) is our specialty, and we delivered a system that the existing Fathym DevOps team could manage and scale on their own.
Results
Server performance increase
Prototype to production
Data points served per month
A 1000x improvement in performance allows a single production server to serve billions of points a month at around 10% capacity. With improved communication and a better understanding of the offering, Fathym now confidently markets and sells the product, and is well positioned to maximize on their investment in weather forecasting and machine learning.
Read