ServicesPricingWorkAboutBlogContact

Schools in rural Tamil Nadu wanted tablet-based learning but without two problems that come with most implementations: hardware that takes days to replace when it breaks, and apps that assume a stable internet connection. Off-the-shelf apps are built for typing-first, always-online classrooms. A Class 7 school in the Nilgiris has intermittent connectivity and students who learn by writing on paper. A cracked screen cannot be allowed to erase a term's work or stop a lesson.

The central design question was handwriting. Students needed to write homework, notes, and full tests by hand on the glass with a stylus — not tap through multiple-choice questions. That meant capturing ink on-device, scoring it with AI, and giving teachers a way to review and adjust marks before anything was final. During exams the app had to stay locked: no notes, no browser, no way out of the test.

The tablets also had to survive children. A student should not be able to uninstall the app, wipe its data, or leave the lesson mid-session. When a tablet broke, the school needed a spare ready the same day with the student's work already on it — which meant the server, not the device, had to hold all the data.

Scholar Leaf runs as an Android tablet app built with React and Capacitor, backed by a Flask service and SQLite database on a cloud VPS. The handwriting canvas draws ruled lines and faint trace text for younger students to follow. When a student submits work, the canvas exports as an image and goes through an AI vision pipeline on OpenRouter, with a fallback model so a provider outage never blocks marking. AI marks first; teachers get a review queue where they adjust the score, add feedback, approve, or send it back for a re-mark.

Exams run in a restricted mode that blocks learning tools and navigation until the student submits. At the OS level the app registers as device owner and runs in lock-task mode, so the tablet is pinned to Scholar Leaf: it cannot be uninstalled, its data cannot be cleared, and the student cannot leave mid-test. Location sharing is opt-in and student-controlled.

We made the cloud the source of truth so each tablet is replaceable. Every submitted assignment, test result, attendance record, note, and profile lives on the server and restores when a student signs in on any device. Schools keep a pool of pre-locked spare tablets — about three per fifty students. When one breaks, they hand over a spare the same day; the student logs in and picks up exactly where they left off. Notes sync two-way with client IDs and revision tracking to resolve conflicts. One deployment covers multiple schools, with live online classes and separate dashboards for students, teachers, admins, and school managers.

ReactFrontend
CapacitorAndroid App
TypeScriptLanguage
FlaskBackend
SQLiteDatabase
OpenRouterAI Gateway
GPT-4.1-miniAI Marking
NVIDIA NemotronAI Fallback
LiveKitLive Classes
DevicePolicyManagerKiosk Lock
GunicornApp Server
NginxReverse Proxy
1:1Student-to-tablet model, works fully offline
~30Modular backend services with unit test coverage
Same-daySpare-tablet swap with zero data loss
4Role dashboards: student, teacher, admin, school
OngoingTotal Duration

Scholar Leaf started as a working prototype and is now a deployed platform on a cloud VPS with a signed Android build. The first phase established the handwriting canvas, AI marking pipeline, and role dashboards. Later work added the written-test marking workflow, OS-level kiosk lock, device-swap cloud restore, multi-school tenancy, and live classes. A validation pilot in a Nilgiris school is the next step before a wider rollout across Tamil Nadu.

Want something similar?

We build systems like this for teams worldwide. Tell us your idea and we will make it real.

Start a conversationView all work