Welcome to Product Cocktail, where the takes are as polarizing as a shot of Fernet—but the insights come together like a perfectly crafted daiquiri.
The Shake
We interrupt your regularly scheduled tokenmaxxing takedowns, vibe coding cage matches, and hot AI goss to bring you... an intimate tour of a highly anticipated, unreleased AI consumer app: Proof. (Don't worry, there will be cocktails served.)
Loyal readers will recall, I'm building an app in public. It's an AI cocktail logging app for people who want to remember, rate, and reconstruct cocktails they have tried at bars.
Where the app sits: I have a real backend, it's connected to a real model, and I have beta versions available on Android (a signed APK file you can sideload) and a TestFlight version on iOS. I've tested it and it works.
The app is about as real as it can be before my old friends at Apple approve it for public consumption in the App Store (s/o to all of the fun we had building HBO Max IAP 😅).
I made a call for testers last week and some of you replied. Here’s what you’re signing up for.
The Cocktail, as promised: The best margarita you will drink this month.
Gather your ingredients. Real simple: any decent tequila, actual limes, agave syrup.
Go find your shaker and ideally a jigger, but a tbsp will work in a pinch. (Yes, even that dusty cobbler shaker with the little "built-in strainer" that constantly-gets-stuck-together-when-any-amount-of-room-temp-to-cold-temperature-variation-hits-it will work.)
Prepare a drinking vessel with lime + salt on the rim (if you're a real one) and a bunch of ice.
Pour the ingredients in a 4:2:1 ratio. 1 tbsp agave (or little less for a more tart marg). 2 tbsp lime juice. 4 tbsp tequila. (If you actually have a jigger, 10 points to Gryffindor: 2oz / 1oz / 0.5oz.)
Add a handful of ice to your shaker, slam the lid on that thang, and shake the absolute bejesus out of it for 10-15 seconds.
Strain it into your glass and enjoy.
Let there be authentication
I know, I know, boring but smooth like a perfectly foamy whiskey sour. (Thanks, Supabase.) You start off by providing an email and receiving a magic link to sign in. No passwords, no bullshit.
Welcome to the party
Start your cocktail journey the same way you started AP Bio: by writing your name on the inside cover of your notebook. Then you're ready to start cocktailing.
Drinkerrrrs, staaaaart yooour catalogs
Two photographs, like the front and the back of a check. We already learned why last month.
Look at this photograph
Start the core flow by capturing the cocktail plate, a square photo of your drink for the catalog.
We already know what's in the box
After you confirm the cocktail photo, you capture the menu and tap-to-select the menu entry (and drag the corners, if needed), featuring the infamous red box, covered in issue 16. The on-device OCR reads the selection as a quick check to make sure the right item is captured in your crop.
Where the AI model earns its keep
The Review stage, with a confidence label on every field the model identified.
The model reads the printed ingredients and estimates the spec and preparation (method). In this case, the model identified an orange twist garnish with high confidence. The menu never mentioned it. That’s where the drink photo comes in.
Testers, this is where you earn your keep. Any field you correct here is recorded in the database along with the original.
Performance Review
To close out your specimen record, you can optionally rate the drink (in coupes out of five, obviously), log how it tasted, and record any notes (not pictured). The palate descriptors are suggested by the model and you can correct if desired.
What a fine specimen
The finished cocktail record, Specimen No. 001, now saved in your log.
The Log
Once you've added a handful of cocktails to the log, the filtering comes in handy. Easy single tap filtering by base spirit, with an expandable row of filters (location, family, rating) and sorts (newest/oldest, top rated/lowest rated).
Known Issues
No offline filing. If you're off the grid, you're SOL. Tester fam, let me know if you run into problems caused by connectivity issues.
No manual entry. Every specimen must be logged in-situ. No ex post facto logging, yet.
Extraction takes time. As I look for ways to optimize the prompt, expect about a 7-12 second lag for the model to process the images before you see Review.
What's next?
I'll run a 2-3 week beta on iOS and Android with the first 10 people who reach out. (Hit me up via email, carrier pigeon, smoke signal, etc.)
You'll get to be the first to the party and your captures will shape the model choice I eventually land on.
While I'm letting the live data come in, I'll be hard at work on some new features:
The Back Bar
Courtesy of letting v0 cook: cocktail reconstruction based on your home bar inventory, with additional bottles identified that will "unlock" additional drinks.
I said it in June and I'll say it again: This is the single biggest pain point I have as a home bartender, and [v0/Opus 4.8] guessed it and solved it.
Account Management
App Store / Play Store age gating and privacy requirements (FUN!)
Exploring new ideas: social / community angles, interactive cocktail ideas
The
No cap, cap.
I funded the Anthropic API for $50 worth of cocktails during the beta, no auto-reload.
Once we hit that cap, it’s kind of a Cinderella app-turns-into-a-potato situation.
If you can power through that budget during the beta (~1000 cocktails)… Well, color me impressed, you little degenerates.
Product Cocktail
Tip Your Bartender
Send me questions, feedback, and cocktail recipes:
[email protected].
Icons made by Icongeek26 from www.flaticon.com.