Run the Industrial Project course on one self-hosted tool
A proposal to run the course on Frappe Framework, an open-source tool, plus a small set of written AI routines. No bespoke platform is built. This site presents the proposal and a working prototype: a fictional semester you can open at any point and log into as any of its people.
What
One self-hosted open-source tool (Frappe Framework) plus written AI routines. Nothing bespoke.Access
Customers, students, mentors and the coordinator each log in and see only what their role allows. Students can never see mentor feedback, scores or the agreed requirements. This is tested, not assumed.AI
Reviews proposals, writes weekly digests, prepares mentors for presentation days and merges their feedback. It never grades.Cost
No licences, hosting on a server the coordinator already has, and an estimated $50-150 of AI usage per semester.What the course needs
- Collecting project proposals from customers and refining them over several rounds.
- Agreeing requirements with each customer while withholding part of them, so students have to elicit them.
- Forming 8-10 teams of 2-3 students from their preferences, with a fair way to handle disputes.
- Following each team weekly without relying on access to the customer's repository, issue board or meetings.
- Briefing every mentor on every project before the presentations, then collecting and merging their feedback.
- Assessing individuals fairly when some students contribute less.
- Protecting proprietary customer material.
The proposal: three parts, no custom application
| Part | What it is | Who maintains it |
|---|---|---|
| The course site | A Frappe Framework installation. It holds every record and is the only interface people use: forms, lists, comment threads, Gantt and kanban views | Configured once; copied for each semester |
| Playbooks | Written instructions, one per routine, that an AI follows ("review a proposal", "write a weekly digest"). Text files in a git repository, in an open format several AI providers read | Edited as text between semesters |
| The scheduler | A timed job on the same server that runs the playbooks with one AI model and writes the results back | A few lines of configuration |
Two small scripts sit beside the playbooks because they must be exact, not AI-generated: team assignment and the confidentiality test.
What the prototype shows
14
checkpoints across a whole semester, from the call for proposals to the closed course. Any one restores in about 20 seconds.50
fictional people: 11 customers, 28 students in 10 teams, 8 mentors, a coordinator, a course manager, and the AI service account.25
planted situations with a known right answer, from a proposal far too large to a handover with no owner.0
leaks found by the confidentiality test, which runs at every checkpoint against every person in the cohort.Take the 15-minute tour or log in yourself
Decisions needed from the department
- Approval to build and demonstrate the prototype.
- A modest AI budget, on the order of $150 per semester.
- Whether student data may be held on the coordinator's server for a pilot, or whether the university can host it.
- Access to the material from earlier runs: the rubric, the grading scheme and past proposals.
- Who acts as course manager.