Your project
One semester, one topic you choose, and a pipeline another person can run.
Every student builds one project, individually, on a topic of their own choosing. The project follows the exercise chapters of the required text: you gather a corpus, turn a concept into measures, analyze the result for a stated purpose, and communicate it so that someone else can check your work. The six milestones are the stages of that one project.
Working on your own corpus means your results differ from everyone else’s. That is the point. In each review you explain how your data shaped your decisions, and you see how a classmate’s data shaped theirs.
Choose a topic
Pick a topic in public affairs, public policy, management, or your own research area that you want to understand better: how a field frames a policy problem, how a management idea spread, who collaborates on an issue, or how the study of a concept has changed. A workable topic has three properties.
- A literature you can retrieve. A first search should return a few hundred to a few thousand records with abstracts. Fewer leaves too little to code and analyze. Many more makes each step slow without teaching you more.
- A concept you can define. Milestone 3 asks you to measure one concept in the text, such as an equity or an efficiency framing, with at least two categories. Choose a topic where such a distinction matters to you.
- A question a reader outside your field could care about. Your final artifact is written for a non-specialist audience, such as a journalist, a policy analyst, or a program manager.
Try Chapter 1’s three basic moves on a candidate topic before you commit: summarize and interrogate one paper, draft a literature search, and ask a model to critique your design. Then run a first query and read ten records. The direction memo in Milestone 1 records what you decided.
Use OpenAlex unless you have a reason not to
The book’s exercises are written for a corpus of scholarly works retrieved from OpenAlex, a free and open index, and the companion website’s examples use it. Build your corpus there unless another source serves your topic better.
You may use a different source with the instructor’s approval, for example agency reports, public comments, or organizational documents. The source has to support the same work:
- records with text you can code and a stable identifier for each;
- a relational field, such as authors or organizations per record, if you plan the network route in Milestone 4, or metadata to predict from if you plan the prediction route;
- terms that let you store, analyze, and describe the records in your submissions.
Ask before Milestone 1 is due. Some exercises name OpenAlex; apply each one to your approved source.
The project, stage by stage
Swipe or scroll the table sideways to read every column. With a keyboard, focus the table and use the arrow keys.
| Stage | Read | Do | What you end with | Milestone |
|---|---|---|---|---|
| Project setup | Chapters 1–2 | Set up the Rules/Memory/Knowledge project structure and draft your direction | A workspace an agent and a reader can follow, and a one-page direction memo | 1 · January 27 |
| Data management | Chapter 3 | Chapter 4 exercises: review the source, retrieve the corpus, diagram it, write the codebook, assess quality, document the workflow | A documented corpus a classmate could rebuild | 2 · February 17 |
| Concept representation | Chapter 5 | Chapter 6 exercises: prepare the coding corpus, build a dictionary, code with a prompt and stress-test it, audit validity and reliability | Two sets of labels for one concept and an audit of the stronger one | 3 · March 10 |
| Data analysis | Chapter 7 | Chapter 8 exercises: a network or a predictive model, one purpose exercise, and a validity statement for the AI-assisted parts | An analysis with a stated purpose, a comparison, and its limits | 4 · April 7 |
| Scientific communication | Chapters 9 and 11 | Chapter 10 exercises: freeze the environment, package the documentation, build the artifact, show uncertainty; then review the ethics of your pipeline | A reproducible project and an artifact for a non-specialist reader | 5 · final session |
| Synthesis | Chapter 12 | Present the pipeline and answer questions | A ten-minute presentation | 6 · final session |
Each stage depends on the one before it. The corpus you document in February is the one you code in March, analyze in April, and communicate in May, so identifiers and boundaries you settle early save work later.
Organize the project the way Chapter 2 describes
Download the project skeleton from the companion website in Week 1 and use it as your project folder all semester. It separates three things an AI agent, and any human reader, needs to find:
- Rules in
AGENTS.md: how work is done here, including what must be verified. - Memory in
memory/: where the project stands (CURRENT.md), what you decided and why (DECISIONS.md), and how your prompts changed (PROMPTLOG.md). - Knowledge in
kb/: what the project knows, including concept definitions, codebooks, and promptbooks.
The same folder is your record of AI use and of the decisions behind each milestone. There is no separate log to keep. Using AI explains what to record and when.
A project you can finish
Scope the project to the six-hour weekly outside-work cap, which also covers reading and reflections.
- Do the required exercises first. The book has 33 exercises; P539 requires a core set and offers a choice within Chapter 8, and the assignments page marks the rest as optional extensions.
- Develop on a small sample. Code 20 documents before 200, and 200 before the full corpus.
- Use the companion’s worked notebooks and sample network data to see a complete example before adapting it, or as a stand-in if part of your own data is not ready.
- Save an unresolved error and bring it to class when troubleshooting would exceed the week’s budget.
Optional extensions do not raise the grading expectations. A narrower corpus with checks you can defend is worth more than a larger one without them.
Feedback changes the work
You receive instructor feedback on every milestone. Before each of Milestones 2 through 5 is due, you also take the work to an in-class exercise review, where a classmate tries to follow or reproduce it and the instructor closes with a worked example.
Keep your first attempt before you revise it. When feedback changes the project, add a dated entry to memory/DECISIONS.md: what you were told, what you changed or kept, and why. Your reflections and the final presentation draw on those entries.
If a problem with data access, tools, or scheduling threatens your progress, contact the instructor early so we can adjust the plan while preserving the learning goals. Use an accessible local or browser workflow; you do not need paid software or a paid AI subscription.
Experiential learning
Experiential Learning (EL) designation: approval pending. The project is designed for Applied experiential learning through project-based learning: sustained work on a question without a predetermined answer, with preparation, mentoring, reflection, and a final product. See the experiential learning activities and evidence →