AI-Assisted Coding Lab

In this lab you will use two AI-powered coding assistants – Claude Code and Google Antigravity – to build a complete data analytics and machine learning pipeline. The goal is to learn how these tools extend beyond simple code generation through features like hooks, skills, subagents, rules, and workflows.

Learning Objectives

• Use an AI coding assistant to plan, build, and test a full ML project
• Configure CLAUDE.md and GEMINI.md to guide AI behavior with project-specific rules
• Create hooks that automate quality checks (linting, syntax validation) on every file change
• Build custom skills that teach the AI reusable workflows (data analysis, model evaluation)
• Use subagents to decompose complex tasks into isolated, parallel work streams
• Compare how Claude Code and Google Antigravity approach the same concepts differently

Problem Statement

Your task is to build a complete ML pipeline for classifying wines into 3 classes using the UCI Wine dataset (sklearn.datasets.load_wine()). The pipeline must include:

1. Exploratory data analysis (EDA) using polars and matplotlib
2. Feature engineering and data preparation for modeling
3. XGBoost classification model training with cross-validation
4. Model evaluation and a comprehensive performance report

You will complete this task twice – once using Claude Code and once using Google Antigravity – to understand how each tool’s customization features support the development workflow. Each part includes a demo/ folder with a reference implementation on the California Housing dataset (regression), so you can see a working example without being able to copy it directly for your classification task.

Prerequisites

• Python 3.11+
• uv package manager installed
• Claude Code CLI installed (for Part 1)
• Google Antigravity IDE installed (for Part 2)

Environment Setup

# Clone the repository
git clone 
cd ai-assisted-coding

# Install uv (Python package manager)
curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env

# Install dependencies with uv
uv sync

# Verify the setup
uv run python -c "from sklearn.datasets import load_wine; print('Dataset OK')"

Lab Structure

Part 1: Claude Code

Located in part1_claude_code/. A single guided lab where you use Claude Code to build the full ML pipeline from scratch. The task naturally exercises every major feature:

• CLAUDE.md: Project-level instructions enforced on every file write
• Hooks: Automated quality checks (ruff, py_compile, test scaffolding) that fire automatically
• Skills: /analyze-data, /evaluate-model, /generate-report for reusable workflows
• Subagents: Explore, Plan, and Bash agents for task decomposition
• Slash commands: /plan for structured planning before building

See part1_claude_code/README.md for the full walkthrough.