Skip to content
Syed Zubair
All work

Case study 02

Generic garden advice fails. This one knows your soil.

A garden planning app that turns a blank yard into a plan you can plant, then tells you what to do each week.

Company
TheHexaTown
Year
2025
Platform
iOS, Android
Stack
Expo, React Native, Supabase, OpenAI, RevenueCat
Status
Live
Three phones with the Dear Garden app, the welcome screen with Petal in front, beside the line: Plan, plant and care for your garden.
46
screens
20
tables, all locked to their owner

01 / The problem

Most garden advice is written for nobody in particular. It ignores your soil, your sun and the climate outside your door. So people buy the wrong plants, watch them die, and decide they are bad at gardening.

02 / What I built

  • A plan for your bed

    Pick a size, a shape and how much sun it gets. The app returns a planting plan with plants placed by zone, quantities and what looks good in each season.

  • Swap one plant

    Do not like a pick? Ask for one replacement in that spot, and keep the rest of the plan.

  • Your conditions, filled in

    A postal code turns into a hardiness zone, a soil type and a soil pH, so you do not have to know any of it.

  • A weekly care list

    Seasonal tasks with a short reason for each one, so the work is spread out instead of dumped in spring.

  • Name that plant

    Point the camera at a plant and get an answer back, then save it to your garden.

  • Credits, not a subscription

    Buy a pack when you want another bed or another design. Nothing recurring to cancel.

03 / Key decisions

01

The planting rules live in the prompt

Zone splits, how much of a bed to fill, plant density, hero plant rules and toxic plant exclusions are written into the request, and the model returns the plan as JSON. That made the rules quick to change, and it means the app has to be careful with what comes back.

02

Credits gate the AI, and the gate is plain data

A count on the bed and a balance on the profile decide whether a generation may run. A purchase tops the balance up. None of it depends on the model, so the paywall keeps working even when the AI does not.

03

The care list is deliberately not AI

Tasks come from a fixed table, copied per user. The weekly list is predictable, it costs nothing to run, and it cannot hallucinate a job that would kill your plants.

04

Every table locked to its owner from day one

All 20 tables use row level security with one ownership rule repeated across the schema, plus a short list of shared plant tables everyone can read.

04 / Screens

Three Dear Garden screens on warm paper: garden styles to pick from, the weekly care list with ticked jobs, and a wishlist of plants from a design
Pick a style, work through the weekly list, and buy from one shopping list.

05 / Outcome

The app is live on the App Store and Google Play.

People start with a blank bed and finish with a plan, a shopping list and a weekly routine.

Get it on the App Store · Get it on Google Play

Contact

Got an idea? Let's build it.

Start a project
Next case studyZero Cheating 03 / 07