AI can write the code. Can your student explain it?
Copilot-era CS education has one honest question left: does the student understand what they submitted? Prova conducts spoken code walkthroughs and technical orals — for every student, at intro-course scale.
Autograders certify that the code works — not that the student can
CS assessment was automated earlier than any other discipline: submit, run tests, score. That pipeline assumed writing passing code required understanding. AI assistants broke the assumption — a student can now produce a correct, idiomatic solution to almost any coursework problem without engaging with it at all.
The industry answer to this exact problem is instructive: professional engineers using AI heavily still do code review, where you defend your changes to another human. The competence being checked isn't typing the code — it's owning it: knowing why it's shaped this way, where it's fragile, what you'd do differently.
Prova brings that review step to coursework. Students submit as usual — autograders still check correctness — and then each student walks through their own code in an adaptive spoken session. Two students with identical submissions can have wildly different understanding; the walkthrough is what tells you which is which.
From CS1 to capstone
Defend your submission
Ten minutes on the student's own code: why this structure, what breaks on edge cases, how it would change under new requirements. Pasted-in code exposes itself immediately.
Theory, spoken
Complexity analysis, data-structure trade-offs, systems design, proofs — students reason aloud through your course material with adaptive follow-ups.
Individual accountability
Each member of a team project defends the shared codebase individually — architecture, their contributions, the parts they didn't write. Free-riding becomes visible.
A career skill, rehearsed
Explaining technical decisions out loud is the skill technical interviews and code reviews demand. Every Prova session is practice for the conversations their careers will be made of.
Fits the pipeline you already have
Keep your assignments, your autograder, and your late policy. Add a Prova walkthrough to the assignments that carry weight: students complete their session within a window after the deadline, and evaluations arrive rubric-aligned with quoted evidence and full transcripts. Review takes minutes per student — a task that distributes cleanly across a TA team, with none of the consistency problems of forty different TAs conducting forty different interviews. Final grades stay with the course staff, and every correction tunes your course's agents to grade more like you do.
Courses typically weight the walkthrough as a multiplier or gate on the assignment grade — passing code that its author can't explain stops earning full credit. Announce that policy on day one and watch how differently students use their AI assistants.
Frequently Asked Questions
What is a code walkthrough exam?
A code walkthrough exam asks the student to explain code they submitted: what it does, why it's structured this way, where it would break, and how they'd extend it. Prova reads the submission, generates questions specific to it, and conducts the walkthrough in spoken conversation — producing a rubric-aligned evaluation with full transcript.
Should CS programs ban AI coding assistants?
Most are concluding they can't and shouldn't — industry has adopted them, and unenforceable bans just push use underground. The workable position is: use the tools, but be able to explain every line you submit. Oral walkthroughs make that standard enforceable, separating students who directed the tool from students who just accepted its output.
How does Prova ask good questions about code?
Prova analyzes the submitted code against the assignment brief and rubric, then probes the decisions in it: why this data structure, what the complexity of this function is, what happens with an empty input, how you'd refactor this if requirements changed. Follow-ups adapt to each answer, so the conversation finds the real depth of understanding.
Does this scale to a 300-student intro course?
Yes — that's the design center. Sessions run concurrently within a window you set, and instructors or TAs review structured evaluations rather than conducting interviews. Intro courses are where AI-generated homework hits hardest and where per-student instructor time is scarcest.
Can it assess theory courses, not just programming?
Yes. Algorithms, systems, databases, theory — anywhere a student should be able to reason aloud about a proof, a design trade-off, or a protocol. Faculty upload their materials and rubric, and Prova conducts the examination on that content.