skills.rest
joaoguirunas
joaoguirunasCOMMUNITY
Explore all skills by this author →
GitHub

ai-ml-data-science

Guides end-to-end ML workflows from EDA and feature engineering to model evaluation and production handoff.

3|3|Updated Apr 23, 2026
Search Tags:#mlops#data science#machine learning#feature engineering#model evaluation#eda#lightgbm
One-click install
npx skills add https://github.com/joaoguirunas/team-os --skill ai-ml-data-science-joaoguirunas
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ai-ml-data-science
Source: https://github.com/joaoguirunas/team-os/tree/main/.claude/skills/ai-ml-data-science
Command: npx skills add https://github.com/joaoguirunas/team-os --skill ai-ml-data-science-joaoguirunas

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) and assets (resource) components.

What problem does it solve? Data science projects often fail from leakage, weak baselines, poor evaluation, and missing production handoff. This Skill provides structured, repeatable workflows that turn raw data and business questions into validated, documented models ready for deployment. ## Core Features & Use Cases - End-to-End DS Lifecycle: Covers problem framing, EDA with drift detection, feature engineering with leakage prevention, baseline-first model selection, and slice-based evaluation. - Production MLOps Patterns: Data contracts, lineage tracking, feature store hygiene, CI/CD/CT/CM pipelines, and production feedback loops with canary deployment. - Ready-Made References & Templates: Eleven operational guides (class imbalance, hyperparameter tuning, interpretability, streaming features) plus copy-paste templates for evaluation reports, model cards, and SQLMesh projects. - Use Case: You receive a churn dataset and need a defensible model. Follow the EDA checklist, build LightGBM baselines, tune with Optuna, run slice analysis, and deliver a model card with monitoring and retraining triggers. ## Quick Start Ask the AI to run an exploratory data analysis and build a baseline classification model on your dataset following this skill's workflow.

Frequently Asked Questions about ai-ml-data-science

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I start a machine learning project with tabular data?▼

Start with a simple baseline such as majority class or linear regression, then train LightGBM as a strong default for tabular data. Escalate to neural networks only when error analysis justifies the added complexity.

How do I prevent data leakage in feature engineering?▼

Ensure every feature is available at prediction time, use time-based or group-based splits instead of random splits for temporal or user-correlated data, and never compute aggregates over windows that include the future.

LightGBM vs XGBoost vs CatBoost: which should I use?▼

LightGBM is the fastest strong default for most tabular tasks. CatBoost often wins when datasets have many high-cardinality categorical features thanks to native ordered target encoding. XGBoost is a mature alternative with similar accuracy.

What metrics should I use for imbalanced classification?▼

Use PR-AUC as the primary metric for imbalanced data, not accuracy or ROC-AUC. Tune the decision threshold via the precision-recall curve and consider F-beta when false negatives and false positives have different business costs.

When should I use SQLMesh for data transformation?▼

Use SQLMesh when building SQL-based feature pipelines with staging, intermediate, and marts layers, especially for incremental models that need testing and audits. For general lakehouse infrastructure, pair it with a data lake platform skill.

Why does my model perform well in training but poorly in production?▼

Common causes are train-serve skew from divergent feature logic, data drift, and leakage during training. Use a shared feature store for both pipelines, monitor distribution drift with PSI or KL divergence, and validate parity with integration tests.

Related Skills

Explore More Data & Analytics Skills→
ancolemanancoleman
503

ingesting-data

Facilitates data ingestion from cloud, APIs, files, and streaming sources into databases.

Community
Advanced
DazlarusDazlarus

ml-ops

Implements model registries, serving APIs, drift detection, and automated retraining pipelines for ML operations.

Community
Advanced
DataZooDEDataZooDE
37

anofox-forecast

Run 32 time series forecasting models with pure SQL in DuckDB.

Official
Advanced
shinnytechshinnytech

rag-implementation

Build RAG systems with vector databases and semantic search.

Official
Advanced