Generative · Education
Content generation systems for education
In education the load is not teaching but preparing the same topic again for different levels and exercise types. Zero-Shot builds systems that generate exercises, exam questions and level variants from course material, with every output tied to the source and passed through teacher approval.
What we build
- 01Exercise and exam question generation mapped to learning objectives
- 02Level variants of the same content
- 03Question quality checks for ambiguity, multiple correct answers and contradictions with the source
- 04Distractor generation and difficulty balance for multiple choice
- 05A teacher approval flow — no output reaches a student unapproved
How it runs
01
Map the objectives
Which content belongs to which objective is made explicit, and every generated question is attached to one.
02
Sample set with teachers
Teachers mark good and bad question examples, so the definition of quality comes from inside the institution.
03
Supervised generation
Generated questions go to teacher approval, and rejected ones are returned to the system with the reason.
04
Connect to assessment
How well questions discriminate between students is measured, and questions that do not are removed from the pool.
What we measure
- Share of questions passing teacher approval first time
- Time to prepare a question set
- Discrimination and difficulty distribution across the pool
- Questions contradicting the source material (target: zero)
- Objective coverage — where questions are still missing
When this is the wrong fit
The system does not assess students, assign grades or make decisions about individuals; it produces content. Setups that work on student data to make personalised decisions are outside this scope.
Common questions
Do generated questions go straight to students?
No. Every question passes teacher approval, and nothing unapproved reaches a student. The system's job is to put a good draft in front of the teacher, not to decide in their place.
What if a question contradicts the source?
That is tracked as its own error type. Every generated question is compared against the source material, and contradictions are removed before they even enter the approval queue. The target is zero, and deviation is reported.
Is student data used?
Not on the generation side — the system works from course material. Anonymous performance data can be used to measure question quality, with the institution's approval and in a form that cannot be traced to an individual.
Other services
Last updated: 2026-09-05
Let's talk about this one
Tell us what you want to build and within two days we'll tell you who takes it and how we'd measure it.