Bayer is expanding their service offering to build further competitive advantages in their industry by empowering farmers through better yield estimations, earlier disease and pest detection and treating their fields more effectively. The overall goal is to leverage machine learning to give their customers better recommendations and build a stronger relationship with them, built on trust. The ultimate aim is to pass on their centralized know-how to their customers worldwide.
The assembled team was solid, with the right mix of skills. However, they had to piece together tooling to develop reliable and scalable processes. The team ended up spending many hours building and maintaining integrations between tools, negotiating licenses and service agreements, and documenting processes. The methods worked well with no surprises, but they struggled to stand up to real-world challenges. Multiple vendors were startups where getting them through the required procurement processes was challenging. The number of vendors aggravated this issue – nobody enjoys paperwork, background checks, legal checks etc.. tasks that no ML engineer or buyer wants to take care. Furthermore, these vendors would update their APIs and other aspects as their tools advanced, forcing the team to update their internal solutions constantly. “We are continually having to implement improvements for unannounced API changes from our vendors; it’s holding us back.”
Where could the team find a comprehensive, enterprise-ready solution that they could depend on to grow their new department?
Hasty.ai helped us speed up our ML workflow by 40%, which is fantastic. It reduced our overall investment by 90% to get high-quality annotations and the resulting model.
Initially, Bayer Crop Sciences started using Hasty for the annotation tool and its automation. The fact that the models used to automate the annotations were learning on the project was particularly beneficial as expert entomologists and botanists had to label this very specific image data. The automated labeling generated by the custom models reduced the time to process an image by a factor of 10. With this automation, the Bayer team could create a model and deploy it in less than a week. The strong results led to the approval for six more follow-on projects. Bayer developed a workflow that allows them to create applications for a broad base of their customers rapidly.
Bayer built an impressive instance segmentation model with only 56 images. This model was already producing acceptable results for the intended use case. With Hasty, the application development team developed and deployed the models directly into their iOS and Android apps without setting up serving infrastructure. The team realized that they could move beyond annotations to model building in Hasty. They are also working with the team to build an inference engine to handle their enormous surge demands, enabling the team to deliver on the ‘real-time AI’ promise.
“With Hasty, we can execute on rapid prototyping for various object detection and segmentation use cases and benefit from an initial glance of the performance of the AI model. The excellent extension capabilities also facilitate the integration of existing data lakes.”
With the apps in production, Hasty enabled Bayer to provide value to the customer from the start. The success of the first projects has given the ML team the green light to extend their offerings and increase their budget significantly for the coming years. Bayer Crop Sciences has established a process and a software setup. They can approach multiple applications confidently, knowing that they will deliver on time, on budget, and reliably.
80% of vision AI teams don’t make it to production because of bad or insufficient data. Hasty solves removes that risk.