Course Content

The course is eight 200-minute sessions: a five-week core course plus a three-session capstone. Each week opens with a 30–60 minute “Before you come to class” checklist and ends with a deliverable due Sunday 23:55.

Core course

Week 1: LLMs, Harnesses & Setup

Starting from the free chat box: what LLMs are, context windows and what happens when you share documents, free vs paid subscriptions and when to upgrade, and the harness idea — coding agents (Claude Code, Codex, GitHub Copilot) that read whole folders, run code, and use tools. Set up a working harness (desktop app + CLI); close with OpenRouter and local open-weights models.

Which AI? See my take on current models.

Content

Week 2: From Raw Data to Report

Download real survey data, build a reproducible pipeline (explore → composite variables → aggregate → join an external API), and contrast an undirected “vibe report” with a carefully directed report.

Case study: World Values Survey

Content

Week 3: Data Wrangling & Debugging

Reviewing AI’s work: project instruction files (CLAUDE.md / agents.md), reusable skills, the three kinds of tests, git discipline, and safe autonomous execution.

Case study: Austrian Hotels

Content

Week 4: Econometrics with AI

AI as a research companion for causal identification — designing control sets (helpful vs adversarial), surfacing instruments via prompt chaining, and difference-in-differences.

Content

Week 5: Text as Data

Turning short texts into numbers — the NLP pipeline, four ways to score sentiment, humans vs AI, and a text→data pipeline that scales via an LLM API.

Case study: Football Manager Interviews

Content

Capstone Project (3 sessions)

Three-session team project on manager changes in football. One big research question, messy real-world data, AI as your primary teammate. — Project description

Session 1: Data collection & description

Pick a league, collect and document 10+ seasons of match and manager-change data.

Content

Session 2: From text to expectations (APIs)

Scrape news sources and use an LLM API to score expectations around each manager change.

Content

Session 3: Difference-in-Differences analysis + final presentation

Run a DiD on your panel, explore heterogeneity (including by expectation score), and present to the class.

Content

--- title: "Course Content" --- The course is eight 200-minute sessions: a five-week core course plus a three-session capstone. Each week opens with a 30–60 minute "Before you come to class" checklist and ends with a deliverable due Sunday 23:55. # Core course ## Week 1: LLMs, Harnesses & Setup Starting from the free chat box: what LLMs are, context windows and what happens when you share documents, free vs paid subscriptions and when to upgrade, and the harness idea — coding agents (Claude Code, Codex, GitHub Copilot) that read whole folders, run code, and use tools. Set up a working harness (desktop app + CLI); close with OpenRouter and local open-weights models. **Which AI?** See [my take on current models](../da-knowledge/which-ai.html). [Content](../unit1/) ## Week 2: From Raw Data to Report Download real survey data, build a reproducible pipeline (explore → composite variables → aggregate → join an external API), and contrast an undirected "vibe report" with a carefully directed report. Case study: [World Values Survey](../case-studies/VWS/) [Content](../unit2/) ## Week 3: Data Wrangling & Debugging Reviewing AI's work: project instruction files (`CLAUDE.md` / `agents.md`), reusable skills, the three kinds of tests, git discipline, and safe autonomous execution. Case study: [Austrian Hotels](../case-studies/austria-hotels/) [Content](../unit3/) ## Week 4: Econometrics with AI AI as a research companion for causal identification — designing control sets (helpful vs adversarial), surfacing instruments via prompt chaining, and difference-in-differences. [Content](../unit4/) ## Week 5: Text as Data Turning short texts into numbers — the NLP pipeline, four ways to score sentiment, humans vs AI, and a text→data pipeline that scales via an LLM API. Case study: [Football Manager Interviews](../case-studies/interviews/) [Content](../unit5/) # Capstone Project (3 sessions) Three-session team project on manager changes in football. One big research question, messy real-world data, AI as your primary teammate. — [Project description](../capstone/) ## Session 1: Data collection & description Pick a league, collect and document 10+ seasons of match and manager-change data. [Content](../project01/) ## Session 2: From text to expectations (APIs) Scrape news sources and use an LLM API to score expectations around each manager change. [Content](../project02/) ## Session 3: Difference-in-Differences analysis + final presentation Run a DiD on your panel, explore heterogeneity (including by expectation score), and present to the class. [Content](../project03/)