All projects
ai agents
A deployed image classification service
Train an image classifier, evaluate it honestly, and serve it behind an API that survives real input.
Why it matters
Machine learning roles screen on evaluation discipline and data leakage far more than on model architecture. A deployed model with an honest confusion matrix answers both, and very few portfolio projects do.
What you build
- A training pipeline with a genuinely held-out split
- Evaluation beyond accuracy, including per-class error
- An inference API that handles unexpected input
- A container that runs the whole thing anywhere
If you want more
- Confidence thresholds with a fallback
- Monitoring for input drift
Build it
Train, validation and test. Peeking at test is how people report 99% and ship 60%.
Material for the whole build
Shows the full model lifecycle with evaluation you can defend — including where the model fails and why.
Get a plan built around projects like this