ACM Calunga

The Problem

The store was tracking stock and sales through manual, spreadsheet-based processes with no real-time visibility, no audit trail, and no way to prevent overselling or restrict sensitive data (like purchase cost) from staff who shouldn't see it.

Key Features

• ▸Ledger + cache consistency design: stock quantity is a denormalized cache kept in sync with an append-only movement ledger, avoiding expensive aggregation while preserving a full audit trail
• ▸Deadlock-safe atomic sales: cart items are sorted before row-locking so concurrent sales sharing products in different order never deadlock; totals are always recalculated server-side, never trusted from the client
• ▸Role-based data and field scoping enforced through Policies and API Resources — cashiers' queries are scoped to their own sales, purchase cost is hidden from the Salesperson role entirely server-side
• ▸Immutable audit logging via a dedicated service called explicitly at write points, capturing old/new values, IP, and user-agent with no edit or delete route ever exposed on the log
• ▸Offline-aware PWA: cache-first for the app shell, network-first with IndexedDB fallback for reads — but offline writes are deliberately blocked to avoid data-integrity risk, a conscious documented trade-off
• ▸Physical inventory reconciliation: stock counts generate adjustment movements automatically from counted-vs-expected differences
• ▸Ops-conscious deployment: isolated Nginx site and its own Postgres database on a shared VPS, with automated daily encrypted-auth backups and 14-day rotation

Tech Stack

Role

Designed and built: architecture, backend API, frontend SPA/PWA, database schema, deployment, and operations (backups, systemd services) — bringing in contractors when the scope called for it.