# Build ML Models With Expert Workflow

ML model projects need a repeatable path from data preparation to deployment. This skill gives Claude, Codex, and Claude Code a structured workflow for training, evaluation, and packaging.

## Install

```bash
npx skillstore add dnyoussef/when-developing-ml-models-use-ml-expert
```

## Metadata

- Status: approved
- Slug: dnyoussef-when-developing-ml-models-use-ml-expert
- Version: 1.0.0
- Author version: 1.0.0
- Skillstore revision: r1
- Version status: valid
- Tree hash: 6dc8610b810c0a039aa5702f173b738d864f8bbf6d01033f5c11b394daffcf6c
- Author: DNYoussef
- GitHub username: DNYoussef
- License: MIT
- Repository: https://github.com/DNYoussef/ai-chrome-extension/tree/main/.claude/skills/machine-learning/when-developing-ml-models-use-ml-expert
- Ref: c579efd02184f1128d6741d16e6c8d951d7d4208
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, filesystem
- Quality score: 72
- Public page: https://skillstore.pages.dev/skills/dnyoussef-when-developing-ml-models-use-ml-expert
- Manifest: https://skillstore.pages.dev/api/skills/dnyoussef-when-developing-ml-models-use-ml-expert/manifest

## Capabilities

- Guides data loading, cleaning, splitting, and normalization for supervised learning tasks.
- Shows model selection logic for classification, regression, clustering, dense networks, and convolutional networks.
- Defines TensorFlow training configuration with callbacks, metrics, batch size, and epoch settings.
- Documents evaluation steps with test metrics, classification reports, and training history summaries.
- Packages model artifacts, preprocessing files, reports, and an inference script for deployment.

## Use Cases

- Prototype a Classifier: Build a first supervised learning model with clean data splits, normalized features, and baseline validation metrics.
- Prepare Deployment Artifacts: Export trained models, save preprocessing assets, and assemble a deployment folder with an inference entry point.
- Compare Model Architectures: Evaluate task complexity, feature count, and data size before selecting a neural network or classical model approach.

## Prompt Templates

### Start a Basic Model

```
Use this skill to build a basic classification model from my dataset. Ask for the target column, metric goal, and deployment format first.
```

### Prepare Training Data

```
Use this skill to clean my tabular data, split train validation test sets, select a model architecture, and define training callbacks.
```

### Package for Deployment

```
Use this skill to design a production ML workflow with evaluation reports, model artifact packaging, and inference script generation.
```

### Audit an ML Pipeline

```
Use this skill to review my ML pipeline for overfitting, leakage, metric choice, artifact safety, and deployment readiness.
```

## Limitations

- Examples assume a local dataset named dataset.csv and a target column.
- It does not provide data privacy controls or dataset governance guidance.
- TensorFlow and scikit-learn examples need adaptation for other frameworks.
- Deployment steps are sample packaging instructions, not production infrastructure automation.

## Best Practices

- Define the target metric and acceptance threshold before training begins.
- Keep train, validation, and test data separated throughout preprocessing and evaluation.
- Record model parameters, training history, data assumptions, and artifact versions.

## Anti Patterns

- Do not evaluate only on training data or tune repeatedly on the test set.
- Do not package model artifacts without documenting preprocessing requirements.
- Do not load serialized artifacts from untrusted sources during inference.

## Security Audit

- Audited at: 2026-07-05T12:43:03.548\+00:00
- Summary: Most static findings are Markdown backtick fences or safe relative artifact writes in SKILL.md, so they are adjudicated as false positives. One semantic issue remains: the generated inference script loads serialized ML artifacts without a trust-boundary warning, which can be unsafe with untrusted files.

## Stats

- Views: 183
- Downloads: 5
- Favorites: 0
- Popularity score: 0