Stays in the country
Processing runs on domestic hardware — in the school itself or in the national data centre. In local mode no pupil data leaves the country.
Slavko · school assistant
Safe for children. Children already use AI; the only question is whose. Slavko is a school assistant under the teacher's supervision: pupils' data stays in the country — on equipment in the school itself or in the national data centre — and answers are checked before they reach a child.
Our minds, our children. It does not grade and does not replace the teacher.
In line with the direction of the national AI Development Strategy 2025–2030: Serbian language, domestic infrastructure, the teacher at the centre.
Why local
AI has already entered the classroom through phones and public services. The decision left to the school is whose equipment it happens on, who sees what a child wrote and who decides what an acceptable answer is.
Processing runs on domestic hardware — in the school itself or in the national data centre. In local mode no pupil data leaves the country.
The teacher sets the goal, chooses the material and sees what the assistant answered. The system is not designed to grade, sort or punish pupils.
Some schools lack a connection fast enough for reliable cloud use. That is why the assistant can run on equipment in the school itself; a fully local mode is proven in the pilot.
When AI affects learning and the monitoring of pupils, the European framework treats it as a high-risk area. That is why human oversight is a founding principle, not an afterthought.
What Slavko does
A good tool does not take thinking away from a person — it gives them more time and better insight. Slavko is conceived as five digital interns who do the routine, while the teacher leads the lesson.
Material, tasks and examples in line with what is taught that day; the teacher sets the frame and checks the quality.
A pupil gets an explanation at their own pace and extra exercises — within the limits the teacher has set. Content aligned to the curriculum by subject is arranged in the pilot.
Voice input and reading aloud run on the school's own hardware, without sending the recording to a third party — help for pupils who find it easier to speak or listen.
Children learn to spot mistakes, ask for sources, ask better questions and use AI as a tool, not as a substitute for understanding.
AI is introduced into school not as a new subject but as content within existing ones; the assistant follows that logic.
How it works
From a question in the classroom to a verified answer, in six steps — on the same core as every Slavko application.
A pupil or teacher types or speaks a question in the school interface. Speech is turned into text on the school's own hardware, not at a third party.
The system establishes who sent the question and opens the job under their name, in their own separate space. It knows how much may be spent and which tools may be run.
Only the necessary context is assembled: the school's rules, teaching material and what has already been recorded for that pupil — with dates.
The model — local or cloud, as the school decides — proposes an answer. Calls to models go through a single point that alone holds the keys and keeps the usage ledger. The proposal is only a candidate.
The candidate is checked; an answer that fails goes back for repair or is discarded. A sensitive topic does not go to the child — it goes to the teacher.
The verified answer arrives as text or Serbian speech. A log remains: who asked, which source was used and what was delivered.
Maturity
A serious school system is not sold on promises. That is why what the core does today is separated from what gets tuned for a specific school, curriculum and rules.
Speech transcription and spoken answers run on the organization's hardware, without a public cloud.
works todayThe model's output is a candidate, not a result: it is checked, and whatever fails goes back for repair or is discarded.
works todayEvery user works in a separate space: a job runs under their name and opens only their data.
works todayThe model, the knowledge base and the communication channel are replaceable; the system is not tied to one vendor.
works todayWorks in Serbian, in Cyrillic and Latin, with the school's documents.
works todayOperation on equipment in the school itself, without a permanent connection — a mode measured in the pilot, on the school's hardware.
proven in the pilotAlignment of content with the curriculum by subject, and rules by age.
proven in the pilotOnly with clear rules, consent and minimal data; connection to school systems and measurement of benefit for the teacher, the pupil and data security.
proven in the pilotThe pilot has criteria agreed in advance: where the data is, which source was used, what happens without the internet and who has the final say.
Sovereignty and trust
The assistant gets only as much context as the task requires. Sensitive data sits in separate systems, split by role and access rights.
Pupils' data, teaching materials and work logs are separated by role; everyone opens only their own.
Every important step leaves a log: who asked, which source was used and what the system delivered.
No independent grading, sorting or punishing of pupils — that stays a human responsibility.
Good Serbian, Cyrillic and domestic context reduce the risk of children receiving unsuitable or mistranslated answers.
Serbia already has a foundation: schools connected to fast internet through the Connected Schools project and a national AI platform in the State Data Centre in Kragujevac. The proposal requires no parallel school system.
Who builds it
Slavko is designed by Dr Ognjen Vučurević — a medical doctor and radiology resident who, since late 2024, has led the AI domain at an IT company in Belgrade: system architecture, specifications, acceptance criteria and the bridging of clinical, business and engineering teams.
The founding logic came from a medical environment, where privacy, verifiability and accuracy are not decoration but a condition of work. The same principle is now being adapted to the school.
The role is architectural: frame the problem, design the system, define the acceptance criteria and prove that it works. Implementation is carried out by agents and teams, to specification and sign-off.
Frequently asked questions
Answers to the questions schools, teachers and parents ask most often.
No. It helps with lesson preparation and practice; the teacher sets the goal, chooses the material and has the final say in every educational process.
On the school's hardware or in the national data centre; in local mode it goes nowhere. If the school decides to use a cloud model, only what the school approves goes outside.
No. Speech transcription and spoken answers run on the school's hardware, so the recording does not go to a third party.
No. The system is not designed to grade, sort or punish pupils on its own, nor to track progress without clear rules and consent.
The answer is checked before delivery; whatever fails goes back for repair or is discarded, and a sensitive topic goes to the teacher, not the child.
The first phase is in production: speech transcription and spoken answers run on the organization's own equipment, without a public cloud. Operation on equipment in the school itself, without a permanent connection, is the second phase — a fully local mode is measured and proven in the pilot, on the school's hardware.
No. The model, the knowledge base and the communication channel are replaceable, and the accumulated knowledge and rules remain because they are kept outside the model.
Collaboration
The proposal is a limited, measurable pilot: a live demonstration, work with teachers, verifiable logs and clear rules of use. Scope, schools and pace are agreed in conversation — by e-mail below.
ognjen.vucurevic@gmail.comOur Slavko. Our knowledge.
One core, several public sites — each for its own topic.