Dmitry Dementiev

Dmitry Dementiev

Co-founder & CEO @ GeoPard Agriculture
#269382310 followers0 following
☀️ 2 day streak
>10,000All time
26 KP

About

Building GeoPard: precision ag analytics and a reasoning AI assistant for agronomic decisions. 7 years turning farm data (satellite, soil, yield, topography) into prescriptions machines can run. Based in Cologne. Ask me anything about VRA, ag data, or why farmers reject black boxes. Off duty: coaching U10 football.

Badges

Tastemaker
Tastemaker
Veteran
Veteran
Gone streaking
Gone streaking

Maker History

Forums

GeoPard Agriculture AI Assistantp/geopard-agriculture-ai-assistant•

2d ago

We publish our AI's exam scores, including the one we failed. Should this be standard?

We build GeoPard AI, a reasoning assistant that plans farm prescriptions (fertilizer, seeding) from field data. Launching here Friday (tomorrow).

Before launch, one decision we made that I'd love this community's take on: we test the assistant on 503 CCA-style agronomy questions (the exam human crop advisors take) and publish every category score. 100% precision ag, 95.8% soil and water, 93.8% crop management, 92.9% nutrient management, and 77.5% pest management, the one we haven't cracked yet.

The question set stays private so it can't leak into training data and inflate future scores. But the results, including the weak one, are public.

Our logic: if an AI recommends what to put on someone's field, they deserve to know exactly where it's strong and where it isn't. Black-box confidence is how you lose farmers forever.

•

23h ago

GeoPard Agriculture AI Assistant - A reasoning agronomist behind every acre

GeoPard AI plans precision ag work the way an agronomist would. Ask one question and it reads your field data (soil tests, yield, satellite, topography), asks about nutrient targets and thresholds, shows every coefficient, and validates the plan against agronomic rules before generating. You approve, the prescription lands on the machine (John Deere, CNH, AGCO). Grounded in your own knowledge base, never trains on your data. Benchmarked on 503 CCA-style questions, all scores published.
View more