Assignments are posted here as the course progresses. All submissions go through Canvas unless otherwise noted.
Weekly deliverables
Each week has a deliverable due before class, and each one builds toward the final project. Most combine a written component with Python work, so plan for several working sessions rather than a single sitting. From Week 3 onward, deliverables require quantitative evidence - a statistical test, an agreement metric, a measured cost, or a computed impact - rather than qualitative description alone. See the schedule for topics and due dates.
All weekly deliverables must include an AI use log - a brief record of what you prompted, what the tool produced, and what decisions you made yourself. See the AI use log guide for format and examples.
Wk 1
~1 page | 30 points | Due before Week 2 begins
Describe a realistic GenAI use case in a future analyst role. Address the task, AI's role, human responsibility, verification, and risk level.
Wk 2
~2 pages + code | 25 points | Due before Week 3 begins
Confirm your dev environment works, then benchmark two tool configurations across 5+ prompts on response time, token usage, projected cost, and rubric-scored quality - and recommend one with the numbers to back it.
Wk 3
4 components | 30 points | Due before Week 4 begins
Clean a 2,000-row messy dataset, build a verified data dictionary with missingness analysis, and write a directed report with a formal methods section, a statistical test, and effect sizes - not just p-values.
Wk 4
3 components | 30 points | Due before Week 5 begins
Find bugs in AI-generated code and compute the exact dollar and percentage impact of each, build a reusable 8-check validation suite, and peer review a classmate's Week 3 notebook both manually and automatically.
Wk 5
3 components | 30 points | Due before Week 6 begins
Use AI's four prompt roles to propose and critique an analysis plan, then implement a segmentation with three model-selection metrics and holdout validation, or a mixed model with AIC/BIC comparison and residual diagnostics.
Wk 6
4 components | 30 points | Due before Week 7 begins
Hand-code 25+ texts and compute inter-annotator κ, compare sklearn and LLM classifiers with per-class precision/recall/F1, build a topic co-occurrence network tested against a null model, and analyze cost at scale.
Wk 7
8 components | 30 points | Due before Week 8 begins
Design a rubric and score three outputs, then measure whether the scores mean anything: human-human and human-LLM κ, prompt sensitivity across repeated runs, and a minimum detectable difference.
Wk 8
~2-3 pages | 30 points | Due before Week 9 begins
Analyze a real AI incident: the failure chain across technical, human, organizational, and governance layers, a quantified harm estimate with subgroup disparity analysis, and a safeguard with a numeric trigger threshold.
Wk 9
~2-3 pages | 30 points | Due before Week 10 begins
A practical governance memo for your final project workflow, scored on a likelihood × impact risk matrix. Every operating rule must carry a measurable threshold tied to a metric you know how to compute.
Final project
Wk 10
6 criteria | 60 points | Due Week 10
Reproducible Python notebook, written summary (3-5 pages), governance memo, 10-minute presentation, and comprehensive AI use log demonstrating all four competencies.
The final project is a reproducible workflow package demonstrating your ability to use, evaluate, and govern GenAI in an applied analytic context, completed in pairs. It is introduced in Week 1 and developed throughout the course - your two individual Week 9 governance memos get reconciled into one for the project, and the Week 4 validation suite carries into the notebook. Presentations are peer-evaluated on four dimensions. See the full rubric for details.
Submission
Submit all work through Canvas by the deadline listed on the assignment.