ai-garage

Evaluate a skill with the GitHub Copilot SDK — measure its triggering accuracy and compare its output quality against a no-skill baseline.

Opinionated data cleaning pipeline builder for tabular datasets. Trigger when the user asks to "clean this data", "fix this dataset", "handle missing values", "remove duplicates", "preprocess this data", "data wrangling", "normalize columns", "fix data types", "data quality", or provides messy data that needs cleaning before analysis or modeling. Also triggers on "prepare this data for modeling", "data preprocessing pipeline". Detects column types, handles missing values with strategy recommendations, normalizes strings, fixes date parsing, deduplicates rows, and outputs clean data plus a cleaning report.

Generate structured, stakeholder-ready data science reports from analysis results, notebooks, metrics, and charts. Trigger when the user asks to "write a report", "summarize my analysis", "create a findings document", "executive summary", "write up results", "document this analysis", "stakeholder report", "present findings", or has completed an analysis and needs to communicate results. Also triggers on "make this notebook into a report", "explain these results to non-technical audience", or "prepare a slide-ready summary".

Perform comprehensive Exploratory Data Analysis on any tabular dataset (CSV, Parquet, Excel, or database table). Trigger when the user asks to "explore this data", "run EDA", "analyze this dataset", "what's in this data", "profile this CSV", "show me distributions", "find patterns in this data", or provides a data file and wants to understand its contents. Also triggers on requests to check data quality, find outliers, or understand feature relationships. Generates a complete EDA notebook or script with visualizations, statistical summaries, and a written findings report.

Lightweight local experiment logging and comparison for ML projects -- no MLflow or W&B required. Trigger when the user asks to "track experiments", "log this run", "compare runs", "experiment log", "which model was best", "experiment history", "save this experiment", "leaderboard", or wants to organize ML experiments without setting up external tracking infrastructure. Also triggers on "what did I try before", "reproduce this run", or "experiment management".

Generate and engineer features from raw tabular data for machine learning. Trigger when the user asks to "engineer features", "create features", "feature engineering", "generate features", "transform features", "feature extraction", "build feature matrix", or mentions lag features, rolling averages, interaction terms, target encoding, cyclical encoding, text features, or date features. Also triggers on "improve model with better features", "what features should I create", or "feature ideas for this data". Supports numeric, categorical, datetime, and text feature engineering patterns.

End-to-end LLM fine-tuning workflow -- dataset preparation, LoRA/QLoRA configuration, training, and evaluation. Trigger when the user asks to "fine-tune a model", "fine-tune an LLM", "train a language model", "LoRA", "QLoRA", "adapt a model", "custom model training", "fine-tune BERT/Llama/Mistral/GPT", "prepare training data for fine-tuning", or wants to customize a foundation model for a specific task. Also triggers on "should I fine-tune or prompt engineer", "dataset format for fine-tuning", "how to fine-tune", "Hugging Face training".

Automated hyperparameter optimization for machine learning models using Optuna or scikit-learn search. Trigger when the user asks to "tune hyperparameters", "optimize model", "find best parameters", "grid search", "random search", "Bayesian optimization", "Optuna", "hyperparameter search", "tune this model", or wants to improve model performance by finding optimal hyperparameters. Also triggers on "what parameters should I use", "sensitivity analysis", or "parameter sweep".

Scaffold and build LLM-powered applications -- RAG pipelines, agentic workflows, structured output chains, tool-calling agents, and chatbot interfaces. Trigger when the user asks to "build a RAG app", "create a chatbot", "build an AI agent", "LLM application", "build with OpenAI/Anthropic/LLM", "retrieval augmented generation", "agentic workflow", "tool calling", "structured output from LLM", "AI assistant", or wants to integrate any LLM API into an application. Also triggers on "chat with my documents", "question answering over docs", "AI-powered [anything]". Supports OpenAI, Anthropic, Google, and open-source models via various SDKs and frameworks.

Generate deployment artifacts for ML models -- FastAPI serving code, Dockerfiles, cloud configs, batch prediction scripts, and monitoring setup. Trigger when the user asks to "deploy this model", "serve this model", "create an API for this model", "productionize", "Dockerize this model", "deploy to AWS/GCP/Azure", "create a prediction endpoint", "model serving", "batch predictions", or wants to take a trained model from notebook to production. Also triggers on "how to deploy", "model API", "Flask/FastAPI wrapper for model".

End-to-end machine learning model training for tabular data -- from baseline to optimized model with proper cross-validation, metric logging, and reproducibility. Trigger when the user asks to "train a model", "build a classifier", "build a regressor", "predict this column", "fit a model", "run ML on this data", "classification task", "regression task", or provides a dataset with a target variable and wants predictions. Also triggers on "compare models", "which algorithm is best", or "baseline model". Supports scikit-learn, XGBoost, LightGBM, and CatBoost.

Design and implement neural network architectures in PyTorch or Keras from task descriptions. Trigger when the user asks to "build a neural network", "design a CNN/RNN/transformer", "deep learning model", "PyTorch model", "Keras model", "neural net for images/text/tabular", "create a training loop", "implement attention", or describes a deep learning task. Also triggers on "what architecture should I use", "training not converging", "model architecture advice", or requests for specific components like "add batch normalization", "implement dropout", "learning rate scheduler".

Create ML pipeline DAGs for data ingestion, preprocessing, training, evaluation, and deployment. Trigger when the user asks to "build a pipeline", "create a workflow", "automate ML training", "orchestrate ML", "Airflow DAG", "Prefect flow", "schedule model retraining", "CI/CD for ML", "MLOps pipeline", or wants to automate any repetitive ML workflow. Also triggers on "data pipeline", "ETL for ML", "automate this process", "production pipeline", "cron job for model training".

Recommend and execute the right statistical test for a given question, with plain-English interpretation. Trigger when the user asks about "A/B test", "hypothesis test", "statistical significance", "p-value", "confidence interval", "is this difference significant", "compare two groups", "correlation test", "chi-squared", "t-test", "ANOVA", "power analysis", "sample size", "Bayesian vs frequentist", "statistical test", "effect size", or any question about whether an observed difference is real or due to chance. Also triggers on "which test should I use" or "analyze this experiment".

Generate publication-quality data visualizations from data and intent description. Trigger when the user asks to "plot this", "visualize", "chart this", "create a graph", "make a dashboard", "show me a heatmap/bar chart/scatter plot/histogram", "beautiful visualization", or provides data and wants to see it visually. Also triggers on "improve this chart", "make this plot better", "presentation-ready chart", or requests for specific chart types. Supports Matplotlib, Seaborn, Plotly, and Altair with Tufte-inspired design principles.

End-to-end git workflow: commit changes, push to branch, and open a PR. Handles both staged and unstaged changes. When files are staged, commits them directly. When nothing is staged, discovers all changed files, groups them logically, and creates progressive commits before pushing and opening a PR. Triggers on: "/ship", "ship these changes", "commit push and PR", "commit the staged stuff and open a PR", "push and PR", "create commits from staged changes and raise a PR", "ship it". Do NOT trigger for plain commit-only requests (/commit) or general git questions.

Extract key, actionable insights from knowledge resources and save as structured markdown files in the knowledge/insights/ folder. Handles YouTube video URLs (youtube.com/watch or youtu.be), web blog/article URLs, local PDF file paths, and remote PDF URLs (including arxiv.org, semanticscholar, etc.). Trigger on phrases like "extract insights from", "summarize this", "key ideas in", "mine this for insights", "what's in this", "give me a digest of", followed by any URL or file path. Also triggers when the user shares a link and wants a quick takeaway without reading the full content. For multiple resources in one request, ask standalone vs. consolidated output before extracting.

Generate a complete project pitch: a timed narration script and a PowerPoint slide deck. Guides you through storytelling framework selection, live industry research, calibrated timing, demo integration, and actual PPTX generation with visual QA.