How a semester runs
Every step below exists in the prototype as records, forms and screens, and is played through by script as the person who would do it.
Before the course
- Proposals. Each customer logs in and fills in a proposal: problem, users, desired outcome, constraints, confidentiality level, weekly availability.
- Review. AI comments on each proposal against a checklist (scope for 2-3 students in one semester, ambiguity, missing information, risk). The coordinator adds comments and holds the conversation. The customer revises in place.
- Brief and sealed specification. Each accepted project ends with a public brief and a sealed specification. Every sealed requirement is tagged: stated in the brief; given when a student asks; found only by probing; or deliberately vague and meant to be challenged.
- Pitches and preferences. Customers present. Each student ranks projects and may privately name people they would rather not work with.
- Team assignment. A solver maximises overall preference under published rules (team size 2-3, avoid lists, a project runs only with at least two students), with a published tie-break. Then a short window for agreed swaps, and an appeal form decided by staff.
Each week
| Who | What |
|---|---|
| Team | One weekly report: planned, done, who did what, blockers (and since when), last customer contact, requirement changes, next week. Milestones, requirements with their sources, a dependency log for anything asked of outsiders. |
| Customer | Signs off the report with one click from an e-mail, or asks for changes. A sign-off means "this matches what happened and what I asked for". It is not a grade. |
| Mentor | Signs off the report or asks for changes, and fills in a two-minute check-in per student (present, engaged, on track, concerns). Picks one reported item for the student to show. |
| AI | A digest per team for its mentor, noting what changed and where sources disagree. |
| Course manager, coordinator | One dashboard: all teams by week, sign-offs, milestones, and a "needs attention" list, with a click through to any team. |
A missing sign-off does not block a team; it shows as "no response" after a few days. Nothing depends on a meeting being recorded or on access to the customer's repository.
Set points
At week 3, mid-semester and the end, AI compares each team's requirements with the sealed specification and reports what was found, missed or invented. Teams must cite a source (a meeting or message) for every requirement.
Presentation days
- Before: every mentor receives a one-page briefing per team, with suggested questions for each student drawn from that student's own reported work.
- During: each mentor fills in the same short scoring form per team. A board shows the order and how many scores each team has received.
- After: AI merges the forms per team and drafts written feedback. The team's mentor edits and approves it, and it goes to the students as a PDF.
Grades
AI assembles a decision sheet per team with every source linked. The team's mentor recommends grades with a short justification; the course manager decides. Because every mentor scores every team, a consistently harsh or generous scorer is visible in a calibration view.