MACHINE LEARNING · AGRITECH
ROLE
AI & Data Analytics
TOOLS
Python · scikit-learn · Pandas
TIMEFRAME
Nov 2023 to May 2024
IMPACT
+15% crop yield across pilot farms
Smallholder farms generate operational data such as planting schedules, inputs, weather exposure and yields, but rarely have anyone to turn it into decisions. FarmChoice set out to change that: could data science produce recommendations a farmer could actually act on?
I built predictive models in Python on operational farm data, testing which combinations of inputs, timing and conditions best explained yield differences between comparable plots. The models generated targeted, plot-level recommendations rather than generic advice, prioritised by expected yield impact. We validated against pilot farms and iterated with real feedback from the field.
The recommendations contributed to an estimated 15% increase in crop yield across pilot farms. For me it was also proof of a transferable truth: the same ML playbook that optimises telecom offers (segment, predict, recommend, measure) works wherever there is operational data and a decision to improve.
BY THE NUMBERS
+15%
estimated crop yield increase
7 mo
from first dataset to field-tested models
1
playbook: segment, predict, recommend, measure
Competitor, pricing & churn intelligence
