Education
A hundred and twenty assignments to mark at the weekend, the same lesson prepared three times by level, and an office typing enrolments and answering the same deadlines every September. Inferana moves that work into an approved tool, with pupil data inside the centre.
What each regulation in your sector requires, and which part of that the deployment solves.
Admission, assessment and exam fraud detection are high-risk; emotion recognition in education is prohibited.
Open models with published model cards and licences, access control and an auditable record of every query to document teacher oversight.
Consent and processing of children's personal data with reinforced safeguards and minimisation.
Prompts and responses are never stored or used to train models, and pupil work stays inside the centre's perimeter.
Digital rights in education and protection of the personal data of children under fourteen.
Access control per user and per team, with a record of who queried what and when, exportable for the data protection officer.
Public schools and universities categorise their systems and apply the corresponding security measures.
On-premise inside your own infrastructure, inheriting the categorisation and the controls the centre already has accredited.
The same ones you already use, specific to your sector.
Nobody signed anything off, but a hundred and twenty assignments are due back on Monday and free tools are one click away. What follows is always the same:
I mark a hundred and twenty assignments at the weekend and grades go in Monday.
Teaching staffI prepare the same lesson three times, one per ability group.
Head of departmentEvery curriculum adaptation is requested in writing and takes me half an afternoon.
Learning support leadDuring enrolment I type records and answer the same deadlines forty times.
Academic registryThe grant paperwork eats the morning I had set aside for research.
Research academicThis is children's data and I have no idea which server it lands on.
Data protection officerIf a mark comes out of a machine, the family will challenge it.
Head of studiesGrading that affects a student falls under Annex III of the AI Act as a high-risk system, and a teacher signs the decision. What we provide is marking support and an auditable record of every query, so the centre can document who reviewed what and stand behind it when a family challenges a mark.
They are never stored or used to train models. The text is processed to produce the answer and that is where it ends. Inference runs on European infrastructure or in the centre's own data centre, so a minor's work stays inside the perimeter the centre controls.
It can draft them. The pupil's report is processed inside the centre's perimeter, with no prompts or responses stored, and the text comes out in the structure you already use. The pedagogical decision and the signature belong to the support department, which knows the pupil and answers for the document.
Flat rate. The cost is known at signature and does not move with the number of queries, so it fits an annual budget line and a procurement file. Per-token pricing means estimating a whole teaching body's usage across a school year, which is exactly what nobody can forecast in September.
We do not offer a generated-text detector. Exam fraud detection is high-risk under Annex III, current detectors are unreliable, and a false positive lands on one specific student. That call needs the centre's own procedure and a teacher's judgement.
Yes, with the same key. There the query does go out to the model provider, so pupil assignments and research data are handled with the open models inside the perimeter. Access is configured per team.
Access control per user and per team, with a record of every query. Management and the data protection officer can review usage and export it. The record is there to answer a challenge and to document usage for an audit.
It gets the same treatment: inference on European infrastructure or on the campus's own, with no prompts or responses stored. The models are open and the API is OpenAI-compatible, so a group can build its pipeline and take it along if it changes provider.
In the demo we go through your case: which models you need, where they run and what it takes to meet the regulation that applies to you.
We reply within one business day.
