Teaching software to read a cornfield, and to tell a grower what to do next.
- Role
- Group Product Manager
- Team
- AI, Data and Ops
- Domain
- Drones, computer vision, agronomy
- Model
- B2B2C SaaS
01 — Context
Taranis captures leaf-level imagery of farmland from the air, at a scale no scout could ever walk.
Its customers, ag retailers and the growers they serve, don’t need more images. They need to know which field needs attention, why, and what to do about it.
02 — Challenge
Turn a stream of aerial imagery into an assistant growers trust. The models had to identify weed species, plant diseases, nutrient deficiencies and insects reliably enough for an agronomist to act on them.
Then the product had to present those findings in a way that saved time instead of creating more work.
03 — Role
I led the product end to end: discovery with agronomists and growers, the roadmap, and delivery across three cross-functional teams of ML engineers, agronomists, designers and developers.
I worked closely with the computer-vision team on what the models should detect, how confident they needed to be, and where a person should stay in the loop.
I also owned growth for the mobile app: push-notification experiments, in-app guidance flows and email workflows that drove acquisition and activation.
“The model’s output isn’t the product. The decision it enables is.”
04 — Process
- 01
Observe
Field research with agronomists and growers to learn how scouting decisions are actually made.
- 02
Define
Translate agronomic needs into detection targets and confidence thresholds the vision team could train toward.
- 03
Guide
In-app guidance that takes a user from a detection to an action, measured on adoption.
05 — Outcomes
92%
Automation rate across the detection workflow
+112%
Mobile engagement from push-notification experiments
+47%
Feature adoption from in-app guidance