Build a Crop-Disease Classifier for a Smallholder Agritech Startup
Overview
What this challenge is about.
You receive a curated 22,000-image cassava-disease dataset across 5 classes (4 diseases + healthy) plus a labeled 1,200-image held-out test set. Train a CNN classifier (start with EfficientNet-B0 or ResNet50, fine-tune from ImageNet). Apply class-balanced sampling, standard augmentations, and reach at least 0.85 macro-F1 on the held-out set. Serve the model behind a FastAPI endpoint returning the top-2 labels and confidences. Document monthly inference cost at 50,000 calls per month.
The Brief
What you'll do, and what you'll demonstrate.
Beat 0.85 macro-F1 on cassava disease classification with a CNN that costs under EUR 200 per month at 50,000 inference calls.
Earning criteria — what you'll demonstrate
- Fine-tune ImageNet-pretrained CNNs on domain-specific classification
- Handle moderately imbalanced multi-class data
- Ship a model behind a real HTTP inference endpoint
- Model inference cost at realistic traffic levels
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.
Computer Vision Engineer
Owning a CNN classifier from data through deployed endpoint is exactly the first quarter of work for a junior CV engineer at any product-AI company.
This challenge sharpens
- image-classification
- cnn-architectures
- transfer-learning
Machine Learning Engineer
Shipping a FastAPI endpoint with cost modeling is core MLE work on small product teams.
This challenge sharpens
- pytorch
- fastapi
- cost-modeling
MLOps Engineer
Quantization and inference-cost modeling at production traffic is bread and butter for MLOps engineers on cost-conscious product orgs.
This challenge sharpens
- fastapi
- cost-modeling
- pytorch