GEMINI.md for Python Data Science Projects

# Project: [PROJECT NAME]
## Data Science Project

## General Instructions
- Python 3.11+ with type hints on all function signatures
- Use numpy docstring format for all public functions
- Prefer pandas for tabular data, polars for performance-critical paths
- Use pathlib.Path instead of string paths
- All data transformations should be reproducible (set random seeds, log parameters)

## Code Organization
- `src/` — source code modules
- `notebooks/` — exploration only, never production code
- `data/raw/` — immutable original data (never modify)
- `data/processed/` — cleaned/transformed data
- `models/` — trained model artifacts
- `tests/` — unit and integration tests

## Data Handling
- Never hardcode file paths — use config or environment variables
- Log data shape and basic stats at each transformation step
- Validate data schema on load (use pandera or pydantic)
- Document all assumptions about data quality

## Modeling
- Always establish a baseline model before complex approaches
- Track experiments with MLflow or Weights & Biases
- Report metrics with confidence intervals, not point estimates
- Include feature importance in model outputs

## Notebooks
- Clear narrative flow: question → data → analysis → conclusion
- Each cell should have a purpose (no "scratch" cells in committed notebooks)
- Use markdown headers to structure the story
- Display data shapes after transformations