Rob Gentul
← All work

Personal build · Home

Tell it how you cook. It plans the shelves.

A pantry planner: save your shelf measurements and what you keep, say what matters ("protein snacks should be easy to reach"), and it proposes where everything goes, in three styles, with a shopping list of bins that actually fit.

Role
Product owner: scope, direction, every approval
When
October 2026
Status
Private beta on my Mac; household trial next
Built with
Claude plans, Codex builds
ReactViteTypeScriptCloudflare WorkersPlaywright
171Unit and server checks

All passing in the latest verification run

63Item types placed

Across 66 shelf and floor spots in the test plan

3Styles per plan

Each with an after image and a printable plan

Counted from the build’s own verification records. No time saved has been measured, so I don’t claim any.

The problem

I wanted to organize a pantry the way a good professional would: from the real shelves, in a few styles to choose from, with the exact products to buy that fit the measured space.

What I owned

I set the scope and direction through a long grilling session and approved the plan. Claude planned and reviewed, and Codex built it.

How it went

Private first, public-ready later

Version one is just for our household, but the data model is built so it could open up later.

Stand-in products, no scraping

I chose stand-in products for v1 instead of scraping retailer images.

A review that said no

The first independent review came back NO-GO with 15 blocking issues. We took 19 of 22 fixes in full, then the recheck said GO.

People type preferences, the app places items

Midway I changed direction: instead of assigning each item to a shelf by hand, you say what you care about and it handles the placements.

How I know it works

  • 171 checks: unit and server checks, all passing.
  • Browser tests: 4 focused cases pass in both Chromium and WebKit at desktop widths.
  • Originals untouched: hash checks confirm the 53 saved inputs never change when it builds a plan.
  • No purchases, ever: it makes a shopping list. It never buys anything, and the beta makes no outside AI calls.

What I’d do differently

The first deeper bin it picked failed the geometry check against the real shelf, so it got swapped for shallower bins that fit. I’d check fit against real dimensions earlier. It still can’t prove physical fit or capacity, and the site says so.

Next

Let’s talk.

Questions about this one? Grab 30 minutes.