Technical peers interested in practical AI and ML model work
AI and ML Model Creation
Creation and training of machine learning models, inference design, and the security boundaries needed before models are exposed to users.
This portfolio area is for hands-on AI and ML work: building models, training them, testing inference paths, and then asking the harder security questions around data, artifacts, runtime boundaries, confidence, failure modes, and evidence.
The point is not to present toy demos as products. The point is to show that I can reason across the full chain: model creation, operating constraints, user experience, and secure exposure.
What This Demonstrates
- Building and training machine learning models.
- Turning model outputs into honest user-facing decisions.
- Designing inference around authentication, upload controls, rate limits, and audit records.
- Separating public explanation from protected execution.
- Knowing when a model should say "I do not know" rather than forcing an answer.
Current Model Evidence
The Cats vs Dogs classifier and Dog Breed stack are available as controlled registered-user inference paths when their feature flags are enabled. The Dog Breed path preserves the local model comparison and weighted-confidence lesson from the old site while deliberately disabling the external AI fallback in this rebuild.
Model evidence
Creation, training, inference, and security boundaries.
Runnable model
Cats and Dogs Inference
A trained image classifier with private previews, latest-run history, upload validation, audit records, and an isolated model service.
Runnable model
Dog Breed Inference Stack
A transfer-learning, confidence-weighting, and fallback-boundary stack using a custom model, EfficientNetB0, and ResNet50 through the controlled inference service.