Leou. runs assignments, mock tests and real examinations for tuition batches — with a face-verified login, code that executes in sealed containers, and feedback that tells a student exactly what to fix.
Each of these is a working surface in the product today, not a roadmap item.
Registration trains a face template. Every subsequent sign-in runs a liveness check and matches against a 512-dimensional L2-normalised ArcFace embedding, decided on the server with a majority vote across samples. Image bytes are never written to disk — only the vector is stored.
SNSAI reads a submitted answer back against the mark scheme and reports what it found, what to fix, and the reference. It is deliberately framed as a study aid: six honest verdicts, and never a final mark — oral and written answers still go to a teacher.
Six languages run in throwaway containers for every submission: Python, Node, Java, C, C++ and Rust, plus PostgreSQL, MySQL and SQLite for database questions. Each one is capped at a single CPU, 256 MB of memory and 256 processes, with the network switched off entirely.
# Read n and print the n-th Fibonacci numberdeffib(n):a,b=0,1for_ inrange(n):a,b=b,a+breturna(fib(10))
Students pay first and prove it after — transaction ID, a screenshot, or a photo of the UPI screen. Tesseract reads the amount, date, time and payee straight off the image and pre-fills the admin's checklist, so verification is a comparison rather than a transcription exercise.
Every type is authored once in the bank and then delivered per batch — with images, explanations, difficulty and its own marking scheme.
Plus difficulty tags, per-subject topics, optional images, worked explanations and hidden test cases with their own weights.
Role-focused surfaces over a single data model — a parent sees a read-only window, never a control panel.
A face-verified login and a workspace per batch.
The whole institute, from one console.
A read-only window onto their child's progress.
Untrusted code runs in a container with the network switched off. Faces are matched on the server against stored embeddings and the images are discarded. Every login, submission and payment decision is written to an activity log with the IP address and a device fingerprint.
Objective answers and code tests are graded the moment a submission lands. Oral and written answers go into a teacher's queue with SNSAI's findings attached as a starting point. Nothing reaches a student until the marks are released.
Face-verified accounts, sealed code sandboxes and honest feedback — ready for your batch today.
Illustrative screens — the showcase above uses invented demo data, never a real student's.