✓ 入选在 12 个提案中排名第 1风投级教育 2026年10月26日

面向大学的规模化口试平台

为院系提供软件与工作流，支持大规模开展简短、结构化的口试（viva），这是目前AI难以伪造的考核形式。

可直接粘贴到 Claude Code、Cursor 或任何编程助手的需求说明（英文）。

I want to build a venture-scale product. Help me plan and start the MVP.

Product: Scalable oral exam platform for universities
What it is: Software and workflow for departments to run short, structured oral exams (vivas) at scale, the one assessment format AI can't fake.
Who pays: Department chairs and teaching & learning centers at universities with large STEM programs, paying $15k–60k/yr per department from academic integrity / instructional budgets.
Closest alternatives: HireVue (hiring interviews, not academic mastery), ExamSoft (written proctored exams), Gradescope (written work grading), manual ad-hoc vivas run by faculty
The gap: Oral assessment is the consensus AI-proof format in this discussion, but no product exists to run it at scale: scheduling hundreds of 10-minute vivas, generating per-student question variants from the actual homework, recording, rubric scoring, and flagging inconsistencies between submitted work and oral answers. Incumbents all target written deliverables.
Why now: In the last 12 months, frontier models made take-home written assessment unverifiable — a Fields medalist publicly abandoning homework grading is the signal; universities tried Turnitin/Rumi detectors and found them inadequate (per the article), so they need a structural alternative, not detection.
Moat to build toward: Rubric and question-variant banks per course compound over time, recordings create grading-consistency data, and integration with LMS/registrar systems creates switching costs; institutional sales create a reference moat.
Biggest risk: Universities are slow, budget-constrained buyers with enormous institutional inertia — the article itself notes the system may not change for years even when broken.

First three steps:
1. Build a viva scheduler + question generator that turns a course's homework set into per-student oral follow-up questions, with recording and rubric scoring.
2. Run a paid pilot with one math/CS department (50–200 students) offering 10-minute recorded vivas as 20% of the course grade.
3. Add an inconsistency report comparing each student's submitted written work to their oral answers to flag likely AI-dependent submissions.

Start by asking me up to five questions about constraints (stack, budget, timeline), then propose the smallest version a paying customer would use, and scaffold it.

总分6.6/10

痛点7

付费意愿6

时机8

市场空白7

护城河6

可行性6

支持

谁会付费

拥有大型STEM项目的高校系主任及教学发展中心，每个院系年付15,000至60,000美元，预算来自学术诚信或教学经费。

讨论中的声音 · 4

文章认为作业本身不应计分，应通过考试加口试来验证掌握程度，因为作弊书面作业如今太容易。查看评论 ↗
文章建议教师随机点名，让学生在课堂上讲解或重现自己提交的作业，以发现没有亲自完成作业的人。查看评论 ↗
文章推荐回归课堂书面或口头考试，将其视为唯一可靠的考核方式。查看评论 ↗
一位兼职教授表示，如果现在重新任教，他只会采用现场考试，或许再加一次面试，这样AI就没有空间虚高成绩。查看评论 ↗

内容转述自讨论，原评论请点击链接查看。

竞争对手与空白

HireVue (hiring interviews, not academic mastery)ExamSoft (written proctored exams)Gradescope (written work grading)manual ad-hoc vivas run by faculty

在这次讨论中，口试被普遍视为抗AI的考核形式，但目前没有产品能支持大规模运行：要安排数百场10分钟的口试、基于实际作业生成每位学生的题目变体、完成录制与评分，并标记提交作业与口头回答之间的不一致。现有产品几乎都只针对书面作业。

为什么是现在

过去12个月内，前沿模型让远程书面作业难以核实。一位菲尔兹奖得主公开放弃批改作业，便是明确信号；文章也指出，大学尝试的Turnitin、Rumi等AI检测工具效果不足，因此需要结构性的替代方案，而不是继续依赖检测。

护城河 · 6/10

每门课程积累的评分标准库与题目变体库会随时间增值；录制内容能形成评分一致性数据；与LMS及教务系统的集成会提高切换成本；机构客户的标杆案例也构成壁垒。

反对

红队: 口试正是各大学普遍趋向的抗AI考核方式，而现有产品中没有能规模化支持口试排期、题目变体生成和评分的方案。机构销售周期很长，但结构性缺口确实存在。
最可能的失败原因: 高校是出了名的慢速买家，预算紧张，制度惯性极大。文章本身也提到，即便现行制度已经失效，它可能数年都不会改变。

起步

最初三步: 开发口试排期与题目生成工具，将一门课程的作业集转化为每位学生的口试追问题目，并支持录制与评分。
与一个数学或计算机系合作开展付费试点（50至200名学生），将10分钟录制口试计为课程成绩的20%。
增加不一致性报告，对比每位学生的书面提交与口头回答，标记可能依赖AI完成的作业。
原始讨论: AI use leads UC Berkeley math professor to see 'no point' in grading homework ↗ 231 点327 条评论源自评论原文链接

本文译自英文。阅读原文

支持

反对

起步

本期其他点子