Working pilot

Parallel oral examinations with a virtual examiner

AI is a problem for take-home work – but a huge help in oral examinations.

  • Many students
    in parallel
  • Teachers stay
    in control
  • Transparent
    and verifiable
Illustration: the Polibeli exam screen during an exam – drawing a topic, preparation, answering the virtual examiner, then the content analysis. Illustration with fictional data; interface translated – exams currently run in Hungarian.

Three roles, one examination system

Student

A safe, focused exam environment.

Illustration: the student’s preparation screen with the drawn topic and the countdown.

Logs in with an assigned ID, draws a topic, prepares, then answers the virtual examiner out loud.

Supervisor

A live overview of the whole room.

Illustration: the supervisor’s room view with students’ exam stages changing live.

Sees where everyone is, how much time is left and who needs help.

Teacher

Structured analysis, retrievable answers.

Illustration: the per-requirement analysis and content coverage reviewed by the teacher.

Approved topics, weighted requirements and analysis backed by quotes.

Illustration with fictional data; interface translated – exams currently run in Hungarian.

Why now?

Take-home work no longer proves knowledge. An oral answer does.

In the age of generative AI it is increasingly hard to tell from a well-written essay whether its author truly understands the concepts, can explain connections and apply what they learned. In an oral exam, students have to account for their knowledge there and then, in their own words.

With large cohorts this would take an enormous amount of teaching time. Polibeli solves the capacity problem: it runs individual exam sessions in parallel, while the institution keeps control of the curriculum, the requirements and the final decision.

Printed pages of a take-home essay on a desk, next to a laptop under evening lamp light

Thirty to forty individual oral exams in one room – each one built on the student’s own line of thought.

Knowledge becomes verifiable when it is spoken.

The Polibeli virtual examiner keeps oral exams personal, fair and professionally reliable even at scale – under teacher control.

  1. 01

    Topic draw

    The student picks from face-down cards; the server assigns the topic at random. One draw only.

  2. 02

    Preparation

    A separately configured preparation time with the topic always visible. Students who are ready can start early.

  3. 03

    Answer

    A self-directed exposition, then targeted follow-up questions. The system records the answer and produces a transcript.

  4. 04

    Teacher’s decision

    The teacher decides on the grade based on the per-requirement analysis, the transcript and the report.

A bright seminar room with rows of students wearing headsets with microphones, each answering at their own laptop

One room, many individual exams

30–40 students at once – each with their own examiner.

Every student sits at their own laptop with headphones and a microphone. The examiner appears on screen as a restrained, continuously looping video; students hear each question in their headphones and read the same text on screen.

30–40
laptops in one exam room
1
supervisor view for the whole room
20–30
topic cards fit on one screen
~2 s
refresh of the supervisor overview
Illustration: the seating plan of a 32-seat exam room where every desk runs its own exam session at its own pace.

Two exam modes

Students choose: with notes or speaking only.

A student in a headset types a short outline on her laptop before answering
+

With notes

During preparation students can type their own outline and reopen it while answering; while speaking they see the recognised text of their answer. The outline is not assessed – the live answer is always spoken.

A student in a headset speaks freely at his laptop, one hand resting on the mouse
+

Speaking only

Mouse and microphone only, with no note field. Students speak to the examiner and move on with a few large, unambiguous buttons. The interface shows that the microphone works and recording is running; the transcript is shown once the exam is closed.

Documented answers

Retrievable evidence, transparent assessment.

For every answer, the question-and-answer transcript, the per-requirement analysis and the event log are available in one place. The audio recording is stored separately on the server.

  • Demonstrated, partially demonstrated, not demonstrated and needs-review items – positive findings come with a quote and a rationale.
  • Weighted content coverage, not an automatically assigned grade.
  • A downloadable, structured PDF report with student and topic data, the analysis, the transcript and the events.
Illustration: a time-stamped transcript with search, and the summary by requirement level.

Teacher control

The AI asks. The teacher decides.

I

Approved topics only

The system can draft a proposal from the source material, but it never approves anything automatically. Only teacher-checked topics can be drawn in a live exam.

II

Weighted requirements

Essential, expected and supplementary knowledge with different weights. Follow-up questions move towards the heavier requirements that are not yet demonstrated.

III

Fair questioning

It does not re-ask knowledge already shown, does not hint at answers, and does not assess accent, speaking pace or confidence.

IV

A fixed yardstick

Each exam session works from the curriculum captured at login: later edits never change the yardstick of an exam that has already started.

Where we are

A working system, introduced responsibly.

Built and working

  • Student, supervisor and teacher interfaces
  • Video examiner and Hungarian examiner voice
  • PostgreSQL storage and PDF report
  • Six sample subjects with 24 topics – replaceable with the institution’s own material

Before institutional roll-out

  • A room test with 30–40 real headsets
  • Per-subject calibration of the assessment
  • Alignment with the institution’s data protection and exam regulations

Frequently asked

What people usually ask first

All questions

How does an exam work with Polibeli?

The student logs in with an assigned ID and starts with a device check: listening to a test sentence, checking the microphone and confirming the recording notice. Then the student draws a topic from face-down cards, prepares and answers out loud. The answer starts with a self-directed exposition, followed by targeted follow-up questions. Closing the exam requires a separate confirmation.

Who asks the questions: a person or a machine?

The virtual examiner asks the questions. It appears on screen as a restrained, continuously looping video; the student hears each question in the headphones and reads the same text on screen. A supervisor is present in the room, and the final grade is decided by the teacher.

Does the system assess accent or confidence?

No. The examiner focuses on content: it does not assess accent, speaking pace or confidence, does not hint at answers, and does not re-ask knowledge that was already properly demonstrated.

Who defines the topics and requirements?

The institution’s teachers. Each topic can have a task, source material, an expected answer and detailed assessment criteria. The system can draft a proposal from the source material, but never approves anything automatically: only teacher-checked topics are used in live exams.

What equipment does an exam room need?

One laptop per student with a supported Chrome or Edge browser, a headset with microphone and a stable network connection. The system runs in the browser and needs no installation. Before institutional roll-out we recommend a room test with 30–40 real headsets.

Get in touch

Request a pilot demo

See how the Polibeli virtual examiner can fit into your university’s processes.

Request a pilot demo