Models Need Data
Our client had already developed a disruptive proprietary machine learning model to optimize resource utilization and maximize yields in commercial agriculture applications. Armed with real, impactful results from their own research farms and paying customers clamoring for insights they set out to develop a commercial product.
But these models need lots of data, and the data needs thorough cleaning and formatting. All of this took weeks or months to process by hand and ultimately ended up in a printed report. The reality was, many of the results from the AI are best delivered in an exploratory visual format. Our client needed to enable their customers to upload massive amounts of data, visualize the data before and after processing, and finally scale out their AI model in the cloud in order to be successful. The primary goal was increasing throughput in order to meet customer demands.
A Platform, an App, and an Infrastructure
Fortunately, the team at Lofty is well versed in the language and tooling of Data Science. We assembled a small, agile product team to tackle 3 major initiatives:
- A platform for the storage and processing of customer data
- A web based application to interface with customers for data collection and insight reporting
- A scalable cloud infrastructure to automate and distribute the data science and machine learning workloads running behind the scenes
We brought our expertise in API development, AWS infrastructures, containerization, and data-driven design together to build an incredible platform that is poised to go very, very big.
The Results Keep Coming
Insights from the Machine Learning model are rendered on 3D satellite maps in the final product
Kubernetes, Django, Docker, deck.gl, vue.js, Celery. If you're not in the business of software development that probably looks like word salad to you, but those are just a small selection of the cutting edge tools all working in concert to deliver high-value insights to the user.
The platform is designed for rapid scalability, and capacity to handle additional customer workloads can be added within minutes. The capacity of the infrastructure to handle larger and larger workloads is virtually unlimited.
Full automation of a 1,000 acre farm has been reduced to less than 30 minutes of work for the customer and just a few hours for our client, with plenty of additional optimizations and full end-to-end automation on the horizon. Previously, this had taken as long as 3-6 weeks. That's a 95-97% reduction in processing time, or a 2400% increase in throughput. Woah.
Needless to say, we're excited to see how things pan out for this project. Our client is just getting started and we have only scratched the surface of what has been enabled through this platform. It's made a huge difference already.