Why oral exams need a different kind of prep than a written defense
A written thesis defense and an oral exam test two different skills, even when they cover the same material. Writing a defense document rewards careful phrasing you can revise before anyone reads it — you can look up a term, fix an awkward sentence, and submit your best version. An oral exam rewards something else entirely: retrieving the right fact under time pressure, explaining it in your own words without a script, and handling a follow-up question you didn't prepare for. Re-reading your notes the night before builds recognition — you'll nod along when you see a term you studied — but recognition and recall are not the same skill, and an oral exam only tests recall.
This distinction is not academic trivia. It's the reason "I read my notes three times" so often fails to translate into a confident oral answer. Recognition memory lights up when you see the right answer next to you; recall memory has to produce the answer with nothing next to it. The gap between the two is exactly where oral-exam anxiety lives — not because you didn't study, but because you rehearsed the wrong skill.
Some Russian universities and education commentators have floated proposals to shift final assessments away from purely written VKR-style diploma defenses and toward oral formats, partly as a response to how easy it has become to generate polished written text with AI tools. As of mid-2026 this remains a discussion and a set of proposals in various institutions rather than one settled, nationwide law — some sources describe an initiative to move away from the "habitual practice" of written diploma and course papers, without a single confirmed national reform text. What's consistent across the discussion is the underlying worry: a written document alone can no longer prove a student actually understands the material, so an oral component becomes the harder-to-fake check. Whatever the exact final rules turn out to be, the practical skill you need — explaining material out loud, from memory, under a few follow-up questions — is the same one active recall practice targets directly.
What the two-bot flow actually does
The flow described on this page uses two separate Telegram bots for two separate jobs, on purpose. SummaryBot is built to condense long source material — a lecture transcript, a YouTube link with captions, or a draft of your own thesis text — into structured notes: the thesis of the material, its key points, and its conclusions. StudyBot is built to generate practice questions and explanations from a topic you name, designed for active-recall self-testing rather than for producing more text to read.
There is no single button that runs both in sequence — you get the notes from SummaryBot, look at what topics and points it surfaced, and then type those same topics into StudyBot to generate practice questions. That's an extra manual step compared to a fully automated pipeline, and we'd rather say so plainly than imply a one-click flow that doesn't exist. In practice the step takes a few seconds: you're copying a topic name or a key point, not retyping the whole lecture.
Why not build one combined tool instead? Because the two jobs genuinely need different behavior. Summarizing rewards faithfully condensing what's already there — SummaryBot's prompts are tuned for accurate compression, with strict output-length contracts so you get a scannable recap rather than a second lecture. Generating practice questions rewards testing understanding, not condensing text — StudyBot's prompts are tuned for question generation, hints, and explanations, learning-first rather than answer-first. Keeping them as separate, specialized bots means each one is actually good at its one job, instead of a single generalist tool that's mediocre at both.
A worked example: lecture excerpt to notes to practice questions
Here's a short, realistic excerpt from an intro genetics lecture, followed by what the two-step flow produces from it.
Lecture excerpt (paraphrased, ~120 words):
"Today we're covering Mendelian inheritance and the exceptions to it. Mendel's laws — segregation and independent assortment — predict clean 3:1 or 9:3:3:1 ratios in the classic pea-plant crosses. But most real traits don't follow those clean ratios, because most traits involve incomplete dominance, codominance, or polygenic inheritance. Incomplete dominance gives you a blended phenotype — think pink flowers from a red and white cross, not a 3:1 split. Codominance gives you both phenotypes expressed simultaneously, like the AB blood type. Polygenic traits, like human height, are controlled by many genes each with a small additive effect, which is why height forms a bell curve in a population rather than a handful of discrete categories."
Structured notes (what SummaryBot produces):
- Thesis: Mendel's laws predict clean inheritance ratios, but most real traits deviate from them through three specific mechanisms.
- Key points: (1) Mendel's laws — segregation, independent assortment — predict 3:1 and 9:3:3:1 ratios in simple crosses. (2) Incomplete dominance blends phenotypes (red × white → pink), breaking the clean ratio. (3) Codominance expresses both phenotypes at once (AB blood type). (4) Polygenic traits (height) involve many genes with small additive effects, producing a bell-curve distribution instead of discrete categories.
- Conclusion: Mendelian ratios are the baseline model; incomplete dominance, codominance, and polygenic inheritance explain the common real-world exceptions to that baseline.
Practice questions (what StudyBot generates from the same topic, with expected-answer cues):
- "Explain the difference between incomplete dominance and codominance, with an example of each." — Expected: blended phenotype vs. both phenotypes expressed; pink flowers vs. AB blood type.
- "Why does human height form a bell curve instead of a small number of discrete height categories?" — Expected: polygenic trait, many genes with small additive effects, not a single gene with two clean alleles.
- "A cross between a red-flowered and white-flowered plant produces all pink offspring. What does this tell you about the inheritance pattern, and how would you distinguish it from codominance in the next generation?" — Expected: incomplete dominance because of a blended (not both-present) phenotype; F2 generation would show 1 red : 2 pink : 1 white, distinguishing it from a codominance ratio.
- "Why don't most real-world traits follow Mendel's clean 3:1 ratio?" — Expected: most traits involve incomplete dominance, codominance, or polygenic inheritance rather than simple dominant/recessive single-gene inheritance.
- "If you were asked to defend why polygenic inheritance doesn't violate Mendel's laws, what would you say?" — Expected: Mendel's laws describe how individual gene alleles segregate and assort; polygenic traits still follow those rules at each individual gene, the additive combination across many genes is what produces the continuous distribution.
Notice what the practice questions are doing: they don't just ask you to restate the notes verbatim. Question 3 makes you apply the concept to a slightly new scenario (a follow-up generation), which is exactly the kind of thing an oral examiner does when they ask "and what would happen if..." after your initial answer. Question 5 asks you to reconcile two ideas that might seem to contradict each other, which is a common oral-exam move to test whether you actually understand the material or just memorized the definitions in order.
How to actually use this before an oral exam
Run the notes-then-practice flow a few days before the exam, not the night before — active recall needs at least one repetition with a day's gap to build durable memory, and cramming the flow into a single evening mostly just re-creates the recognition-vs-recall gap this whole approach is meant to avoid. A workable schedule: get your structured notes from SummaryBot as soon as the lecture or draft is available, generate the first round of practice questions in StudyBot within a day or two, answer them out loud (actually speak, don't just think the answer — oral exams test spoken explanation, and silently "knowing" the answer in your head is a different skill from producing it verbally), then repeat the weakest questions two or three days later.
If your oral exam allows or expects you to explain your own thesis or project work rather than a lecture topic, the same flow still applies: paste your own draft or outline into SummaryBot to get the condensed thesis and key points, then generate practice questions from those points in StudyBot. This turns the exercise into rehearsing your own defense — anticipating the questions an examiner is likely to ask about your own work, based on what you actually claimed in it, rather than guessing blind.
One more honest note: this flow is built for self-testing and preparation, not for generating text to submit as your own written work. If a reform in your country does move toward requiring AI-use disclosure for submitted work, using an AI tool to draft your actual diploma or thesis text would need to be disclosed under whatever the final rules turn out to require — using it to generate practice questions you answer yourself, out loud, in your own words, is a fundamentally different and lower-risk use, closer to flashcards than to ghostwriting.