I've spent the last few months building EduVetta, a small SaaS product for teachers. I want to write about what it does and why, without the usual "we're revolutionizing education" framing, because honestly it's a fairly narrow tool solving a fairly narrow problem.
The problem, as I understood it
Teachers spend a lot of time on tasks that are adjacent to teaching but aren't teaching: writing up what was covered in class, formatting a test, chasing down students who missed a lecture, marking a stack of submissions. None of this is hard, exactly. It's just repetitive and it eats time that could go elsewhere.
I talked to a handful of teachers before building anything, and the recurring complaint wasn't "I need AI in my classroom." It was closer to "I keep doing the same clerical task every week and it's tedious."
What we built
EduVetta has four main pieces, and they're deliberately simple:
Lecture and board notes. A teacher records a lecture or uses a digital board. The platform generates a summary and saves the board notes as a PDF, then sends both to the class Telegram channel. This replaces manually writing a recap message after every class.
Question paper generation. You give it a source, a topic description, a YouTube link, an uploaded PDF, or a spoken request, and it drafts a paper with marks and an answer key. You still review and edit before publishing. It's a first draft generator, not an autopilot.
Assignments. One link, no student account needed. Submissions come back in a list rather than scattered across email or WhatsApp.
Quiz games. Same input options as the question generator, but output as something playable rather than a static paper.
Where it's genuinely useful vs where it isn't
Objective grading (multiple choice, fill in the blank) is instant, which is the clearest win. Subjective answers still need a human, and we didn't try to automate that judgment call. I think that's the right line to draw for now, not because the model couldn't attempt it, but because grading written reasoning is exactly the kind of task where a wrong automated answer does real damage to a student.
Students don't need an app or login, they enter a name, roll number and batch. That was a deliberate choice to lower friction, even though it means less structured data on our end.
What I'd still call unfinished
The voice-to-question-paper flow works better for some subjects than others. Anything heavy on diagrams or equations needs more manual cleanup than plain text topics. We're also still figuring out the right defaults for quiz difficulty so teachers aren't stuck tuning settings every time.
If you want to see the actual product rather than take my word for it, there's a free tier with no card required: eduvetta.com Happy to answer questions about the stack or the AI generation pipeline in the comments if anyone's curious.