Team Practicum: Build a Crop-Disease Classifier with a Field Partner
Overview
What this challenge is about.
You receive a labeled dataset of about 8,000 phone photos plus around 1,200 unlabeled photos from a held-out county. Audit and clean the labels (expect 5-10% noise), train a MobileNet-class model in PyTorch, evaluate on the held-out region, and stress-test against the most common failure modes (rain on lens, low light, leaf occlusion). Export the model to ONNX or TensorFlow Lite, document inference latency on a mid-range Android phone (or documented proxy), and write a 4-page handoff doc covering accuracy by class, expected on-device latency, and a list of known failure modes the Android engineer should warn the farmer about.
The Brief
What you'll do, and what you'll demonstrate.
Train, evaluate, and package a maize-disease classifier that an Android engineer can ship to farmers in the field.
Earning criteria — what you'll demonstrate
- Apply transfer learning on a small, noisy real-world dataset
- Evaluate generalization across regions, not just random splits
- Export PyTorch models to mobile-friendly formats and measure latency
- Hand off an ML artifact to a non-ML engineering teammate
Program Fit
Where this fits in your program.
Sharpens the same skills your degree expects you to demonstrate.
Skills
Skills you'll demonstrate.
Each one shows up on your verified credential.
Careers
Roles this prepares you for.
Real titles. Real skill bridges. Pick the one closest to your trajectory.
Machine Learning Engineer
Shipping a transfer-learning model on noisy field data, exporting it for mobile, and writing the handoff doc is exactly the end-to-end MLE work that climate-tech and edge-AI startups hire for.
This challenge sharpens
- transfer-learning
- pytorch
- model-export
Computer Vision Engineer
Real-world image classification with held-out-region evaluation and failure-mode galleries is the bread-and-butter of junior CV engineers at vertical AI startups.
This challenge sharpens
- transfer-learning
- model-evaluation
- edge-deployment
Applied AI Scientist
Diagnosing label noise and quantifying region-shift generalization is the kind of applied-science rigor that startups expect from their first applied-AI hire.
This challenge sharpens
- label-cleaning
- model-evaluation
- transfer-learning