Purven Bhavsar.Power · Systems · Data
← Portfolio

Upasthiti

A mobile-first attendance PWA I built for an NGO, designed to survive patchy field signal and to stop depending on one person's personal spreadsheet.

Role: Architect & Sole Developer

An NGO I worked with tracked event attendance through one person's personal Excel process, a single point of failure that had already become unreachable once. Field volunteers marking attendance at real events often have weak phone signal and shouldn't need to log in just to check someone in.

I designed three different access models for three real roles instead of forcing everyone through the same login: Admin gets full control from a laptop, Viewer is a real login for read-only staff, and Management, the field volunteers, gets no login at all. Instead they get an event-scoped link plus a short fallback code that expires automatically. That last one broke from how I usually build access control, on purpose, because the actual situation (different people rotating through event duty, no time for account creation, working from someone else's borrowed phone) didn't fit a normal login model.

Two real bugs came out of testing this under actual field conditions, both found by stopping the server and reading the database directly rather than just reading the code:

  • The first version of the offline sync only updated the screen from inside a button's click handler. So a change that synced later, after a page reload, left the screen showing stale data even though the database was already correct. I fixed it by making the display always derive itself from the current sync state at render time, so a click and a reload go through the same logic instead of two paths that could drift apart.
  • Registering someone while offline, then restarting the server, created two identical rows instead of one. The cause was React's development mode firing mount effects twice, combined with nothing stopping two sync attempts on the same item from running at once. I fixed it with an explicit lock checked before either the app state or the network call, and confirmed it by repeating the exact sequence (register offline, restart the server) until the database consistently showed one row.

The result is a tested, three-role system with real offline resilience: attendance queues locally when the signal drops and syncs once it's back, with visible "syncing" or "failed, tap to retry" feedback instead of silently losing data.

What this one taught me is when to break from a standard pattern instead of forcing every problem into the same shape, and how to find a timing-dependent bug that only shows up under real conditions (offline, a server restart, React's own quirks) by testing the actual failure instead of just reading the code.

Outcome

Built & tested; not yet on a permanent domain

More work

Want to talk about work like this?

Get in touch →