Rolecaller is an offline-first attendance and data platform for the Malto community's tribal schools in the hills of Jharkhand, India. Geography, language barriers, and thin infrastructure have cut these schools off from consistent educational support for a long time. Internet is a luxury and electricity comes and goes, so a tool built for a connected world just doesn't run there.
The project started with a visit. I spent time with the Malto community in 2023 and saw the gaps firsthand, and at some point I realized that just remembering what I'd seen was passive and didn't change anything by itself. A conversation with my father, where he gently pushed on what I actually planned to do about it, turned that into RoleCaller: an app that lets teachers record attendance on a device that assumes the network doesn't exist, and treats a connection as a bonus when one shows up.
Without a reliable attendance record, a student's absence is basically invisible. Kids in these communities often walk dangerous paths for hours to get to school, and miss days for family obligations, illness, or seasonal work, but when attendance is scattered across paper registers nobody sees the pattern until the kid has effectively dropped out. The cause is systemic instability, not a lack of motivation, and there was no data trail to prove that, step in early, or make the case for scholarships and keeping kids in school.
Community leaders were also very clear about what the tool couldn't turn into: 'We do not want to become the police.' If attendance tracking started to feel like surveillance, or like outsiders imposing their own idea of productivity, it would wreck the trust the whole thing depends on.
- Has to run fully offline. Teachers can go days without a signal, so no core workflow can depend on a live connection.
- Runs on low-end, sub-$100 Android phones in places where electricity is sporadic, so there's no room for a heavy runtime or background work.
- Teachers might get to sync once a day at best, so when a connection does show up, the sync has to be fast and can never lose a day's work.
- Has to support teacher judgment instead of enforcing compliance, which means giving teachers and coordinators visibility without turning into surveillance of students or staff.
Local-first SQLite with a custom "Pulse" sync engine
Each teacher's device treats its local Expo SQLite database as the single source of truth. A custom sync engine waits for a stable network connection, pushes unsynced attendance records to a Node.js API backed by Postgres (Neon), and pulls fresh roster data back down. Pushed records aren't marked as synced locally until the server confirms a 200 OK, and the API upserts on conflict, so a crash mid-sync or a duplicated request can't corrupt or lose anything. Pulling roster updates does a 'smart wipe' that refreshes students and classes while strictly preserving any attendance a teacher hasn't synced yet.
Reason:
Teachers can't be blocked by the network, and a rare, flaky connection window is the only chance to sync, so every push has to be atomic and every pull has to protect whatever work is already sitting on the device.
Tradeoff:
The push/pull protocol and smart-rehydration logic are a lot more code to maintain than "just sync everything," but just syncing everything can silently drop a day of attendance.
UUID v4 identifiers across every entity
Schools, classes, students, and attendance records all use UUID v4 primary keys, generated on the device when the record is created instead of assigned by a central server.
Reason:
A device with no connectivity still needs to create valid, permanent records for new students or classes. UUIDs let it do that without ever asking a server for the next ID, and they slot into Postgres later without collisions.
Tradeoff:
UUIDs are bigger and less readable than auto-incrementing integers, and the schema and API layer need some care to index and join on them efficiently.
Raw Node.js HTTP API with Drizzle ORM over Neon Postgres
The backend skips Express/NestJS for a raw Node.js HTTP server with a shared connection pool, and uses Drizzle ORM to generate type-safe SQL with effectively zero runtime overhead.
Reason:
When a bunch of devices reconnect at once, the sync burst has to be handled quickly on modest server hardware, and for that, cutting framework overhead and cold starts mattered more than developer convenience.
Tradeoff:
No framework means no routing, middleware, or validation for free, so all of that boilerplate gets written and maintained by hand.
Offline-capable authentication
Online logins authenticate against the API and cache a hashed "session snapshot" locally. If a teacher opens the app with no signal, RoleCaller verifies their credentials against that cached, hashed record instead of failing closed.
Reason:
A teacher who hikes to a remote school needs to log in and keep working for days without ever seeing a signal tower. An app that needs a live login check would be useless there.
Tradeoff:
Caching credentials on the device, even hashed, expands the local attack surface and needs more careful handling than a stateless, always-online auth flow.
- Got board approval from FMPB in January 2026 to start deploying across Malto community schools.
- The rollout is staged on purpose: RCPSC and the smaller schools first, then the larger centers once there's trust built up.
- Built a working local-first architecture (SQLite on the edge, Postgres at the core, a custom sync engine in between) that lets a teacher record attendance for days with zero connectivity and sync losslessly on reconnect.
- Designed it with community leaders, who were explicit that they didn't want a "policing" tool, so it stays supportive instead of coercive.
- Knowing about the problem in 2023 didn't help a single student. The useful (and much harder) part was turning it into something a teacher could actually open on a phone.
- 'We do not want to become the police' ended up shaping the app as much as the missing network did.
- Offline-first has to be assumed in every layer from the start. IDs, auth, and sync all break if you bolt it on later, which is why even the primary keys get generated on the phone.
- Starting with RCPSC and the small schools is slower than scaling right away, and the temptation to scale right away is real, but this is a tool that affects actual kids and teachers, so slow is the right call.



