Skip to content

I built a personal drink database using Apple Shortcuts, Google Forms, and Claude

Writing  ✺  AI

A chaotic, trial-and-error journey into automating something nobody asked me to automate.

I have a problem: I go to a boba or coffee shop, I order something incredible, and I customize it perfectly (less ice, 50% sweetness, oat milk). Three months later, I'm standing at the same counter with absolutely no idea what I ordered last time.

So I built a system to fix that.

The Idea

The vision was simple: photograph the drink label that comes with my order, have my phone automatically parse it, and log it to a database I can query later.

The full pipeline looks like this:

  1. Apple Shortcuts captures a photo of the drink label
  2. Apple Intelligence (OCR + AI parsing) extracts the drink details from the label
  3. The parsed data gets submitted to a Google Form
  4. Google Form responses flow into a Google Sheet that serves as my database
  5. Claude reads the sheet and answers questions like "what did I order at Da Vien?"

Simple in theory. Deeply humbling in practice.

Building the Shortcut

I built the Shortcut step by step with Claude's help. The basic flow was:

What sounds like six steps was actually about 40 actions in Shortcuts. And every single one had a gotcha.

Honestly, building in Shortcuts felt like giving instructions to someone who technically does exactly what you say but somehow always misunderstands you.

The Rating System

I also added a ❤️ rating to each log. The Shortcut shows a menu:

Each selection sets a variable to a number (5 through 1) which then gets submitted with the rest of the form data.

Connecting Claude to the Sheet

The last piece was connecting Claude to my Google Sheet so I could just ask it questions in plain language.

My first entry:

Da Vien — Strawberry Cream Oat Latte No ice, 50% sweetness, cream oat milk Notes: 50% sweetness was perfect! Rating: ❤️❤️❤️❤️❤️

Now I can ask Claude "what did I order at Da Vien?" and it reads the sheet and tells me exactly what I got and how I customized it.

What I'd Do Differently

I'd spend more time upfront testing the AI parsing prompt with real label photos before building the whole Shortcut around it. The OCR output from real drink labels is messy and the prompt needs to be robust enough to handle all of it.

Was It Worth It?

Absolutely. Not because it's the most efficient system in the world, but because now I'll never forget that Da Vien's Strawberry at 50% sweetness with cream oat milk is a perfect drink.

And honestly? Building it was fun. There's something satisfying about stitching together a bunch of tools that weren't designed to talk to each other and making them work anyway.

Next

How to manage when processes aren’t ideal

Auto tinting text in Figma