Cursor 2-Day AI Hackathon — Repo Template

---

Sub-Flow (Biokiste)

An adaptive subscription management UI that feels like texting a friend, not navigating a webshop.

Tech Stack

• Frontend: Next.js 16 (App Router), React 19, TypeScript 5
• Styling: Tailwind CSS v4, DaisyUI 5
• Database: Supabase (PostgreSQL)
• AI/ML: Gemini 2.5 Flash (image generation), Vertex AI (A2UI generation)
• Hosting: Vercel

How to Run

# Clone the repo
git clone https://github.com/your-team/sub-flow.git
cd sub-flow

# Install dependencies
pnpm install

# Set up environment variables
cp .env.example .env
# Add your Supabase and Google AI API keys to .env

# Run the development server
pnpm dev

Open http://localhost:3000 to see the app.

Details

The Problem

Subscription management for physical products (veggie boxes, diapers, coffee) is broken. Every service builds an e-commerce shop first, then bolts subscriptions on top. The result: interfaces designed for browsing, not managing.

The Solution

An adaptive subscription interface that:

1. Knows your context — Urgent deadline? Minimal UI. Plenty of time? Let's explore.
2. Learns your preferences — "You always swap fennel. Want me to just do that?"
3. Delights with visuals — AI-generated images of your box and recipes
4. Feels like a conversation — Not a shop, not a dashboard, something warmer

Key Features

• Time-adaptive UI: Same app looks completely different based on urgency (< 12h: urgent mode, > 72h: browse mode)
• Smart suggestions: Pattern detection learns from swap history and proactively recommends
• AI box images: Beautiful food photography generated on confirmation via Gemini
• Persona simulation: Dev panel for instant switching between demo users and time contexts

Demo Scenario

The prototype simulates weekly vegetable box subscriptions with three personas:

Persona	Name	Story	Default Box	Swap History & Patterns
New	Sarah	Just signed up, exploring the service for the first time	Carrot, Broccoli, Apple, Pear, Banana	None — clean slate, no learned patterns
Regular	Mark	3 weeks in, has clear preferences	Fennel, Broccoli, Apple, Orange, Strawberry	Swapped Fennel → Zucchini 3 times; system detects dislike of Fennel (95%) and preference for Zucchini (85%)
Power	Lisa	Engaged power user, curious and exploratory	Carrot, Fennel, Beetroot, Pear, Strawberry	Swapped to Strawberry 2x, Orange 2x, Broccoli away 2x; prefers Strawberry (70%), prefers Orange (70%), dislikes Broccoli (80%)

Things to Try

Scenario 1: Watch a box go through its lifecycle

1. Select Sarah with Plenty of time — explore the relaxed browse mode
2. Switch to Locking soon — see the UI shift to urgency mode
3. Switch to Locked — the box is finalized, no more changes

Scenario 2: Compare "locking soon" across personas

1. Switch to Lisa with Plenty of time and reset boxes — see how a power user's experience differs with learned preferences
2. Select Mark with Locking soon — notice his Fennel → Zucchini suggestion (strong pattern)
3. Switch to Lisa with Locking soon — compare her different preference suggestions (fruit-focused)

Scenario 3: Conversation interactions Try these via the chat interface:

• Add a vacation (pause deliveries)
• Skip this week
• Add or remove items from the box
• Confirm a box, keep editing a box

Architecture

See docs/project-plan.md for the full product vision and docs/phases/ for implementation details across 8 phases.