Building Pharmaciano-ERP: A Multi-Vendor Pharmacy ERP with AI-Based Forecasting

Overview

Pharmaciano-ERP started from a real, observed problem: pharmacy branches I'd seen were still managing stock on spreadsheets, leading to frequent stockouts of essential medicine and overstocking of slow-moving items. The goal wasn't to build another CRUD dashboard — it was to close the gap between "the data exists" and "someone actually acts on it."

Architecture

The app follows a modular, feature-sliced structure on top of Next.js's App Router, separating inventory, forecasting, and vendor modules into independently testable domains. Server Components handle data-heavy dashboard reads directly, keeping the client bundle lean for the lower-end, in-store hardware pharmacy staff often use. Zustand covers the small amount of genuinely global client state (active branch context, cart), rather than reaching for a heavier state library the app doesn't need.

Key Challenges

Keeping inventory counts accurate under concurrent updates. Multiple branches writing to shared stock data at once caused real drift. The fix was moving stock mutations to atomic database transactions with optimistic UI rollback on conflict, rather than trusting client-side state alone.

Dashboard queries slowing down as historical sales data grew. Solved with materialized views for the forecasting inputs, refreshed on a schedule instead of computed live on every request.

Making the forecasting model useful on a limited dataset. Accuracy improves as more branches onboard — this is an area still actively being refined, not a solved problem.

Current Status

Pharmaciano-ERP is in active testing, not yet a finished product. The core forecasting → restocking loop already works end to end, but multi-currency support and offline-first sync for low-connectivity branches are still on the roadmap.

Lessons So Far

The hardest part wasn't the UI — it was designing the forecasting pipeline's data flow so it stayed accurate as branches with different sales patterns were added, without requiring a full model retrain each time. Validating the data model against how real pharmacy staff actually work, early, mattered more than any individual technical decision.