AI Accelerator: Analytics and Data Analysis

Tutorial-style lessons for AI-assisted analytics from exploration to dashboard.

How To Read This Site

Start with the first section, then progress through the remaining modules in order.

The learning material starts notebook-first with Codex in a JupyterLab terminal, then transitions to broader CLI/project workflows for reproducible dashboard delivery.

Pre-Workshop: Set Up Your Environment and Coding Assistant

Setup page: Open Setup

Page	Questions	Objectives
Set Up	What do I need to complete before Session 1 so my machine is workshop-ready?	Verify core tooling: Python 3.11+, venv, pip, Node.js, npm, and VS Code Check that the required tools are installed and working (Python, Node.js, npm, and VS Code) Install Codex and the connector needed for the workshop Sign in to Codex using the method for your Duke affiliation Set up your project environment, install the required packages, and run the final setup checklist

Session 1: Foundations for AI-Assisted Analytics Work

Section page: Open Session 1

Use Codex as a coding agent for analytics work while keeping the analysis understandable, reviewable, and traceable.

Page	Questions	Objectives
Coding Agents for Analytics Work	What can a coding agent do, and what must its output satisfy?	Identify the understandable, reviewable, and traceable requirements for agent-produced analysis.
Prompting and Project Rules	How do prompts and an AGENTS.md file determine what Codex produces?	Write prompts and project rules that produce reproducible notebook artifacts.
Understanding the Data Before Analysis	What does one row represent, and which file answers which question?	Investigate data structure before plotting or interpreting trends.
Planning and Running an Analysis	How have physical and digital checkouts changed over time, and how can Codex help plan a new visualization?	Select the correct dataset, account for partial years, compare visualization approaches, and trace results to code.
Managing Context in Coding Agent Conversations	What belongs in the conversation, and what belongs in project files?	Distinguish working context from saved project context and decide where verified facts, decisions, and workflow rules should live.
Homework: Extend the Keyword Analysis	How can a messy keyword field be investigated with more than one model or method?	Extend the subject-tag analysis, compare approaches, and identify what should remain human-reviewed.
Session 1 Q&A	What questions came up during Session 1?	Review answers to participant questions after the session.

Session 2: Context Management and Agent Skills

Section page: Open Session 2

Choose working file formats that reduce unnecessary agent tool use, then package repeated instructions as reusable agent skills.

Page	Questions	Objectives
Context Management Part 2	How does the working file format change what Codex has to do?	Explain why notebooks can create extra tool use and choose a better working artifact for agent-assisted analysis.
Agent Skills	How can reusable instructions make agent workflows more consistent?	Distinguish prompts, project rules, and skills, then create or evaluate a small analytics skill.

Session 3: From Analysis to Dashboard

Section page: Open Session 3

Use multi-agent workflows to prepare reproducible analysis outputs, preserve provenance, and turn CSV files into a stable contract for dashboard delivery.

Page	Questions	Objectives
Building Datasets with Codex	How do you turn a public data source into a usable teaching dataset? When should you use complete aggregate files or sampled detail files? How do sampling choices shape the claims learners can make?	Investigate a public API access pattern before retrieval. Separate volume questions from composition questions. Define row meaning, sampling strategy, reproducibility, and validation checks.
Multi-Agent Workflows	What is a multi-agent or subagent workflow? When should you use subagents? How do you use subagents without losing control of the project?	Distinguish parent, worker, and checker responsibilities. Decide when a task is ready to delegate. Write bounded subagent task contracts and review the results.
Build and Deploy a Static Dashboard: Prompt Notes	How can Codex turn dashboard-ready CSV files into a static dashboard? How can prompts change the chart, library, or dashboard component? What steps publish a static dashboard with GitLab Pages?	Use prompts to generate and revise a static HTML/CSS/JavaScript dashboard. Keep dashboard data loaded from CSV files at runtime. Follow the basic GitLab Pages deployment sequence for static assets.
Open Dataset Starting Points	What public datasets could support a student dashboard? Which sources have fields that support filtering, aggregation, maps, or time series? How can you choose a dataset that leads to a clear dashboard question?	Compare several public data sources for dashboard potential. Select a dataset with a usable unit of analysis and meaningful dimensions. Turn a dataset idea into a focused dashboard question.