Project Innovation Engine

Project Innovation Engine

Goal

Generate computing project or exam ideas that are:

• Original — not trivially found on GitHub, LeetCode, or standard textbooks
• Feasible — completable by the target cohort within the stated duration
• Pedagogically rich — exercise multiple learning outcomes, not just one

When to use

• When mission.json → evaluation_mode is project and the instructor has no topic yet.
• When a machine exam needs a fresh scenario that students cannot Google.
• When a fil rouge project must span multiple milestones across a semester.

Inputs

• mission.json — level, domain, duration, learning_outcomes, constraints
• Optionally: list of topics already used in previous years (to avoid repetition)

Output contract

project-ideas.md format per concept

## Concept N — <Title>

**Domain:** <sector>
**Difficulty:** <level> — estimated <X>h per student
**Novel element:** <what makes this hard to find pre-made>

### Description
<2-3 paragraph narrative>

### Learning outcomes covered
- [outcome from mission.json]
- ...

### Milestones (if project spans multiple sessions)
1. <milestone 1> — <deliverable> — <week>
2. ...

### Anti-cheat properties
<Why copying from the internet is hard: unique scenario, custom data format, specific API constraints, etc.>

### Risks
<Known difficulties that could block students; suggested mitigation>

Novelty strategies

Use one or more of the following to generate original scenarios:

• Cross-domain combination: embed a computing problem in an unusual domain (e.g. scheduling problem framed as a space mission constraint solver)
• Inverted role: students implement the grader, not the solution (meta-programming, AST analysis)
• Custom protocol: invent a simple domain-specific binary or text protocol students must parse and generate
• Constrained resources: add a novel constraint (memory cap, no external libraries, fixed number of system calls)
• Multi-agent coordination: require two student programs to cooperate via IPC or network

Rules

• Propose at least one concept that can be completed in the stated duration with a 20 % safety margin.
• Flag any concept where the reference solution would take more than 4 hours for an expert — it is likely too complex.
• Do not propose concepts that require licences, proprietary tools, or hardware students may not have.
• Each concept must map its tasks to specific learning_outcomes from mission.json.