Harukibox

The problem

Studios run their flow across 5+ tools: Sheets for costs, LINE for orders, Excel for inventory, paper notes for warehouse, Word for shipping lists. One order crosses four tools — break any link, redo the order. That fractured flow is the ceiling on scale.

What I built

A single system covering paste-list-in to swipe-confirmed-out — end to end.

Paste-list, batch-create
AI parses raw chat orders into buyer, item, price, spec.
Cross-platform alias unification
IG / FB / LINE handles auto-merged into one buyer profile.
Auto cost & shipping prorating
Real-time FX + international shipping prorated per item.
Scan to list
Barcode or photo into AI-parsed product entry in 3s.
Three-tier warehouse
Region → Position → Box. Every item has real coordinates.
Mercari-style card browsing
Visual-first product list. Big photos, clear data.
Swipe-to-confirm links
Per-order confirmation URL. Buyer swipes; done.
AI sales & inventory analytics
Gemini surfaces best-sellers, restock signals, turnover.

Technical decisions

• Supabase + RLS：Account isolation is non-negotiable for B2B SaaS. PostgreSQL RLS is the cleanest answer.
• Google Gemini 而非 GPT-4：Structured output is rock-steady, batch parsing is cheap, and the JSON-schema mode rarely hallucinates.
• Mercari-style card grid:It's the visual language proxy buyers already know. Photo-first, data underneath.

Pricing

What I learned

A narrow niche is the moat. Generic ERPs will never natively solve per-item shipping prorate, cross-platform alias merge, or swipe-to-confirm. One niche solved deeply beats five niches solved 80%.