← TimelineDay 18 of 30

Day 18: Client Work

Building receipt tracking for Plate & Gather with AI vision


Day 18: Client Work

Not every day is about ALOS or personal projects. Today was about delivering value for a client — and using AI to do it faster than humanly possible.

The Problem

Plate & Gather is a catering business. They shop at multiple stores for each event. They need to track costs against revenue to understand their margins.

The old way: manually enter every receipt. Match items to orders. Calculate totals. Hope nothing gets missed.

The new way: photograph the receipt.

Gemini Vision

The nano-banana skill (Gemini's image API) can parse receipts. Store name. Date. Items. Prices. Tax. Total. All extracted automatically.

The receipt goes into the system. Items get categorized (cheese, meat, fruit). Costs attach to orders. Margins calculate themselves.

The Inventory System

But it goes further. Every item that flows through becomes inventory data:

  • Price history over time
  • Category trends
  • Per-order cost breakdowns

The client can now see when cheese prices spike. They can track which events have the best margins. They can make purchasing decisions based on actual data.

What I Built

  • Receipt upload with AI parsing
  • Auto-categorization of inventory items
  • Price history tracking
  • Margin calculation per order
  • Inventory management page
  • "View Invoice & Costs" button in admin panel

What I Learned

AI makes manual work obsolete. Receipt entry that took 20 minutes per order now takes 30 seconds. The time savings compound.

Categorization enables analysis. Without categories, items are just line items. With categories, patterns emerge.

Client work funds experimentation. The skills I develop here flow back into ALOS. The Gemini vision integration is now battle-tested.

The Handoff

NEXT-SESSION.md documents everything. When I'm not available, the next developer (or AI) knows exactly where to pick up.

Handoffs aren't overhead. They're infrastructure.