The Center for Data Innovation recently spoke with Mark DeSantis, CEO of Bloomfield, a Pittsburgh-based company plant imaging technology, about how their AI-powered solutions are transforming agriculture. DeSantis explained how Bloomfield combines deep learning models with custom-built cameras to deliver precise, plant-level insights to farmers.
David Kertai: What does Bloomfield do?
Mark DeSantis: We’re an agricultural technology company, now part of Kubota Corporation, that assesses the health and performance of specialty crops one plant at a time. When most people think of agriculture, they picture row crops like wheat, rice, and corn, which account for about 85 to 90 percent of global crop production. However, there’s another vital category, high-value crops such as berries, cacao, and coffee beans. These crops grow in dense canopies and require frequent, close-up inspections to monitor their health and development. Since satellites and drones can’t effectively penetrate the foliage, accurate assessment relies on capturing ground-level images.
We developed an AI-powered device with a custom imaging system that mounts onto existing farm vehicles, such as ATVs or tractors, that growers already use. This device captures detailed, plant-level information on the condition of each individual crop, whether it’s a grapevine, blueberry bush, or apple tree.
Kertai: How did you design your imaging system to handle the challenges of real-world field conditions?
DeSantis: To generate reliable insights, we need high-quality images, so we designed a custom dual-lens stereo camera. The two lenses provide depth perception, while the camera’s built-in lighting system helps cancel out variable field conditions like sunlight, shade, time of day, or cloud cover. As the vehicle moves, the system captures multiple high-resolution images per second, which are then processed on a local server deployed on the farm.
The AI model analyzes these images to detect and track key plant features such as grapes, clusters, leaves, and shoots, even when they are partially hidden or overlapping in the foreground or background. It continuously monitors individual plants over time, measuring changes in size, shape, and color. For instance, it might detect that the grapes on vine six, row seven, have grown and darkened since last month, helping farmers make more timely, targeted decisions about care and harvest.
Kertai: How does your AI system adapt to new crops, growing conditions, or feature requests from growers?
DeSantis: The system evolves through continuous refinement and by layering in new types of analytics. It detects patterns in plants and improves over time as it learns from large datasets. While each new crop or variety requires some retraining, the model becomes better at adapting as it builds a broader understanding of visual cues.
Growers occasionally request detection of new features, such as specific diseases or mold, which may require either a new model or a significant update. When growers raise new detection needs, we retrain the AI model to handle the added complexity. This flexibility allows the system to scale while also being tailored to each grower’s unique challenges. Feedback from growers directly shapes both our technical roadmap and how we prioritize new deployments..
Kertai: What challenges have you faced as an AI innovator in agriculture?
DeSantis: One of the biggest challenges was gaining farmers’ trust in our system’s accuracy. Early on, they would compare our results to their manual counts—what they called “ground truthing”—and sometimes claimed our data was off. In one case, a grower said our counts didn’t match those of their 20-person crew. So, we suggested, “Let’s ground-truth your ground-truthers.” We double-checked both their numbers and ours, and it turned out their manual counts varied widely. The issue wasn’t with the AI system, it was the inconsistency of the human data. That was a turning point. It reminded us that human benchmarks often show more variability than the systems designed to match or exceed them.
Another challenge is the agricultural calendar itself. In tech, you can run experiments and iterate constantly. In farming, you might only get one growing cycle per year. That really limits how fast you can learn and improve. To work around that we expanded to South America, where the growing seasons are opposite. That gave us two cycles per year, one in the Northern Hemisphere and one in the Southern, accelerating development and validation.
Kertai: What sets Bloomfield apart from others in the agri-tech space?
DeSantis: We focused on one specific problem, providing growers with accurate, plant-level insights into health and performance. This narrow focus was key to our success. Instead of chasing other opportunities like robotic harvesting, we honed in on what growers truly needed and worked tirelessly to perfect it. Additionally, we made sure to go where the crops were. Rather than limiting ourselves to places like Southern California or the Finger Lakes, we expanded to France, Chile, Peru, Morocco, Mexico, wherever berries and grapes are grown. It was risky and expensive, but we believed in meeting customers on their terms, not ours.
