Yumindo CRM - Curtain Production Operations | Arsyadam's Portfolio

Overview

Yumindo manufactures and installs custom curtains — a single order can have dozens of windows with different fabrics, models, and sizes per room. Production progress was hard to track in real time; coordinators had to ask each stage manually (cutting, sewing, finishing, QC), and field survey data often reached the system late.

I built Yumindo CRM, an Indonesian-language web dashboard that manages transactions end-to-end: fabric/model master data, field measurements (AI-integrated), payments, warehouse, and a per-stage production board optimized for coordinators updating status from their phones — with optimistic UI so taps reflect instantly without waiting on the server.

My Role

Frontend / UX: Built the mobile-first production page (Start button on the left for left-handed use, full-width Finish at the bottom), spotlight onboarding tutorials per deploy, queue search & filters, and per-tab caching so switching subtabs doesn't refetch everything.
Backend / Data: Server actions for assign/start/finish production, schema migrations on windows (status, assignee, audit log via window_production_logs), and per-stage queue queries with FIFO queue position.
AI integration: /api/ai-extract route calling Mastra's survey-workflow from the measurement page — Supabase auth plus transaction validation before extraction runs.
Operations modules: Transactions, field ops (survey/installation), warehouse procurement/inventory, master data, and role-based dashboards for CEO/CS/production staff.

Technical Highlights

Production board with optimistic updates for floor coordinators

Production coordinators don't have time to wait on a spinner every time they tap Start or Finish — they're walking the shop floor one-handed on a phone.

I implemented optimistic patches in per-tab client cache: on button press, windows state updates immediately (or the item drops from the queue when it advances to the next stage) while startWindowProduction / finishWindowProduction server actions run in the background. On failure, a toast fires and data resyncs from Supabase. Tab counts refresh async without blocking the UI — cutting perceived latency from ~1–2 seconds per action to instant.

Per-window production queue, not per order

One transaction can have many windows at different stages. I modeled the queue at windows.production_status (PENDING → CUTTING → SEWING → FINISHING → QC → READY) with stage_entered_at for FIFO ordering, per-window assignees, and ASSIGN / START / FINISH audit logs. Each card shows client name, room, and fabric + dimensions — without address or transaction codes cluttering a small screen.

Impact & Results

6 production subtabs (overview, fabric order, cutting, sewing, finishing, QC) with live queue count badges
Coordinators can assign staff, start, and finish a stage from one card — layout tested for left-handed mobile use
Spotlight tutorials auto-run for new users and on each deploy version bump (coach marks on real UI, not generic popups)
Replaced manual/spreadsheet production tracking for Yumindo's internal operations (pilot)

Stack & Architecture

Layer	Tech	Why
Web app	Next.js 15 App Router	Server actions + RSC; one codebase for the ops dashboard
UI	shadcn/ui + Tailwind	Consistent components; dark-first internal design system
Auth & DB	Supabase (PostgreSQL + Auth + Storage)	RLS, survey photos in storage
ORM (read/admin)	Prisma	Type-safe queries for admin tooling
AI bridge	Mastra API (`MASTRA_API_URL`)	CRM doesn't embed models — calls the separate `yumindo-ai` workflow service
Notifications	Novu	In-app notifications for operational events

Lessons Learned

I initially refetched all tab data + tab counts on every subtab switch — on mobile that felt painfully slow. Per-tab caching plus background prefetch fixed it; prettier skeletons wouldn't have. For onboarding, dialog popups didn't stick; spotlight coach marks pointing at real page elements worked in one pass for coordinators.

Yumindo CRM - Curtain Production Operations