AI oral defense, explained
An AI oral defense is a short voice exam after a written submission, and the mechanism is what makes it hold up.
The questions are never shown in advance. They are generated live from the student’s own work, the student answers by voice in two to four minutes, and Tolus scores only what is actually said. The output is a stored transcript and a citable rubric verdict, not a probability.
- Pre-reveal
- None. The defense is live, so it measures understanding under genuine assessment pressure.
- Partial credit
- A wrong answer with real reasoning still scores. A rehearsed, vague one does not.
- Transcript
- Every defense is stored, re-readable, and citable per sub-score in the gradebook.
The defense, step by step
Follow one submission through the whole sequence, from upload to gradebook.
- 01
The student submits.
Any written assignment: essay, lab write-up, problem set, history response. Submitted through Google Classroom or pasted directly.
- 02
Tolus reads the submission and prepares the questions.
The AI extracts the key claims and likely weak spots, then generates three to five voice questions that probe paraphrase, evidence, and transfer.
- 03
The student takes the defense.
Two to four minutes. The student answers each question by voice. Tolus listens, follows up where reasoning is shallow, and stops at the rubric depth the teacher set.
- 04
Tolus scores only what was said.
No gap-filling, no hints. Each rubric category is scored from a quoted moment in the transcript. The model never tells the student the right answer.
- 05
The teacher receives the report.
A weighted score, the full transcript, citations per sub-score, and a flag if the defense looks contested. The score writes back into the Google Classroom gradebook.
Oral defense FAQ
Common questions about how it actually runs.
What is AI oral defense?
AI oral defense is a short, voice-based exam that runs after a written submission. The AI generates follow-up questions from the student's own work, the student answers conversationally, and the AI scores only what the student explicitly said.
Why voice instead of text?
Voice removes the easiest cheating shortcut: copy-pasting a model's answer. It also keeps the cognitive load on the student and produces a transcript that is auditable and citable. The student speaks; Tolus types.
How are questions generated?
Each question is generated from the student's actual submission, not from a fixed template. Tolus probes the specific claims in the work, looks for paraphrase that avoids the original wording, and tests transfer to unprompted examples.
Can the student see the questions in advance?
No. The defense is live. Pre-revealing questions would defeat the purpose of measuring understanding under genuine assessment pressure.
What happens if the student gets a question wrong?
The score reflects what the student said. A wrong answer that demonstrates real reasoning can still earn partial credit; a vague or rehearsed answer does not. The rubric belongs to the teacher.
Is the transcript saved?
Yes. Every defense produces a stored transcript that the teacher can re-read, share with parents, or attach to the gradebook for audit.
See the mechanism on your own work.
We will run a real oral defense on one of your assignments in 30 minutes, then walk through the transcript and the rubric score line by line.