PopUp

A one-tap pop-up store generator for creators, makers, and micro-entrepreneurs selling at weekend markets, craft fairs, food festivals, and pop-up events.

Comment

What Problem PopUp Solves and Why I Built It

Every weekend, 50 million creators, makers, and micro-entrepreneurs set up stalls at farmers markets, craft fairs, and pop-up events. They arrive with beautiful handmade products candles, pottery, jewelry, baked goods and a fatal problem: no way to sell them professionally.

PopUp is a one-tap pop-up store generator. One photo of your products. One minute. One complete digital store with QR payments, inventory tracking, and a marketing video. No hardware. No setup time. No missed sales.

How I Structured Conversations with MeDo to Build PopUp

I built PopUp entirely through natural language conversations with MeDo. Here is how I structured the build process:

Phase 1: Architecture Design

I started by defining the 5-agent system. I told MeDo: "Build a multi-agent pop-up store generator with a Vision Agent that detects products from photos, a Catalog Agent that generates descriptions and pricing, a Checkout Agent that creates QR payment links, an Inventory Agent that tracks stock in real-time, and an Analytics Agent that produces business intelligence."

MeDo generated:

  • The database schema (users, stores, products, transactions, agent_logs, analytics)
  • The agent communication protocol (Router Agent coordinates all 5 agents)
  • The state machine: DETECTED → CATALOGED → ACTIVE → RESERVED → SOLD → OUT_OF_STOCK

Phase 2: Core Flow

I iterated on the photo-to-store pipeline. The first version detected products but did not handle overlapping items well. I refined: "When the Vision Agent sees 3 identical candles, count them as quantity instead of 3 separate products." MeDo updated the detection logic.

I then added:

  • Confidence scoring system: flag detections below 70% for manual review
  • Duplicate detection: identical products counted as quantity, similar but distinct products separated as variants
  • Manual product entry for missed items

Phase 3: Payment and Voice (Turns 23 to 32)

I integrated Stripe for QR code payments and Baidu Voice for hands-free checkout. The breakthrough was voice-activated dynamic QR generation.

How it works:

  1. Seller says "Charge $18 for the blue mug"
  2. Text-to-Speech captures the audio
  3. Large Language Model extracts intent: product identifier and price
  4. System matches spoken description to the nearest product in catalog
  5. Dynamic QR code generates for that exact amount
  6. Seller shows QR to customer
  7. Customer scans and pays
  8. Voice confirmation plays: "Payment received. Eighteen dollars."
  9. Inventory updates automatically
  10. SMS receipt sends to customer

The seller never touches their phone during the transaction. MeDo built this by combining Text-to-Speech capture, Large Language Model intent extraction, and Stripe Payment Link generation.

Phase 4: Intelligence Layer (Turns 33 to 40)

I added the Analytics Agent with real-time dashboards and the Kling video generation for marketing content.

The "Sales Recap Video" feature — auto-generating a 15-second animated video from event data — was entirely MeDo's creation. I described the concept and MeDo built:

  • Kling Text-to-Video integration
  • Format selectors for Instagram Stories (9:16), TikTok (9:16), X/Twitter (16:9), Instagram Feed (1:1)
  • AI-generated caption suggestions with platform-specific tone
  • One-tap share buttons

Phase 5: Polish and Scale

I added:

  • Gamification: Pop-Up Battles, XP levels, streaks, badges, challenges
  • Offline transaction queuing via IndexedDB for markets with no wifi
  • Multi-currency support with ExchangeRate API
  • Bulk import from CSV/XLSX files
  • External integrations: Shopify, Square, WooCommerce
  • Team dashboards with role-based access (Owner, Seller, Viewer)
  • In-app messaging between vendors and customers
  • Customer loyalty program with points and rewards
  • Social media auto-posting to Instagram, TikTok, X/Twitter
  • Custom branding: logo upload, brand colors, custom URL slug

Each feature was built through iterative conversation: I describe the need, MeDo generates, I test, I refine.

The Most Impressive Feature MeDo Helped Me Create

The 60-Second Store Transformation is the feature that makes jaws drop.

A user uploads one photograph of products arranged on a table. Within 60 seconds, MeDo's multi-agent system performs 5 parallel operations:

  1. Vision Agent detects and counts every item, handling overlapping objects, identical quantities, and low-confidence flagging
  2. Catalog Agent generates professional names, descriptions, and market-competitive pricing by cross-referencing Etsy, Amazon Handmade, and local market data via Web Search and AI Search
  3. Kling Image Generation creates clean, white-background product photos from cluttered snapshots
  4. Checkout Agent generates Stripe QR payment codes for every product
  5. Kling Text-to-Video produces a 15-second promotional trailer with native audio and animated charts for social media

The user goes from "I have stuff on a table" to "I have a professional store, payment system, and marketing video" in under a minute. No typing product names. No setting up Stripe manually. No hiring a photographer or video editor.

The second most impressive feature is Voice-Activated Checkout. During a sale, the vendor never touches their phone. They speak: "Charge eighteen for the blue mug." Text-to-Speech captures the audio. A Large Language Model extracts intent and matches to the nearest product. A dynamic QR code appears. The customer scans and pays. Voice confirmation plays: "Payment received. Eighteen dollars." Inventory updates automatically. An SMS receipt sends to the customer. All hands-free.

The third most impressive feature is the Sales Recap Video. At the end of an event, Kling automatically generates a 15-second animated video showing: revenue counter animating up, product photos flying in, "Sold Out!" text appearing, upbeat music playing. It is auto-formatted for Instagram Stories, TikTok, X/Twitter, and Instagram Feed. The user taps once to share. Every video is a free advertisement for PopUp.

How I Used Plugins or API Integrations to Extend Functionality

PopUp integrates 12 official MeDo skills and 4 custom skills I built via Skill Creator.

Official Skills Used

Skill How I Used It
Image Recognition Powers the Vision Agent's product detection from photographs. Detects distinct items, separates overlapping objects, identifies categories (candle, jewelry, pottery, baked good, clothing), estimates attributes (size, color, material), assigns confidence scores, and counts quantities.
Kling Image Generation Transforms cluttered snapshots into professional product photos with clean backgrounds, adjusted lighting, and consistent styling suitable for the payment page and social media.
Kling Text-to-Video (Omni) Generates 15-second marketing trailers with native audio, animated charts, and music. Auto-formats for Instagram Stories (9:16), TikTok (9:16), X/Twitter (16:9), and Instagram Feed (1:1).
Stripe Payments Creates QR payment links for every product. Processes Apple Pay, Google Pay, and card transactions. Handles refunds, tax calculation by geolocation, and discount rules (time-based and quantity-based).
Text-to-Speech Enables voice command capture during hands-free checkout. Provides voice confirmations after payment ("Payment received. Eighteen dollars."). Reads protocols aloud for accessibility.
SMS / Notifications Sends digital receipts to customers. Sends low-stock alerts to vendors. Sends parent/guardian alerts for under-18 sellers. Sends push notifications for new messages and battle updates.
Baidu Maps Shows "Nearby Pop-Ups" for customer discovery. Provides local pricing intelligence for competitive pricing suggestions. Shows event locations and directions.
Web Search Powers Catalog Agent's competitive pricing research. Enables Scout Agent's market trend monitoring. Finds latest research on product categories and safety data.
AI Search Generates product descriptions using trending language from the creator economy. Finds comparable products for pricing intelligence.
Google Text Translation Translates the entire UI and customer-facing store into 20+ languages. Enables multilingual voice checkout.
Login Handles role-based authentication: Vendor, Customer, Parent/Guardian, Team Member (Owner, Seller, Viewer). Supports OAuth and phone number login.
File Storage Stores product photos, AI-enhanced product images, receipt PDFs, sales recap videos, bulk import files, and custom branding assets.

Custom Skills I Built via Skill Creator

ProductTaxonomy

A domain-specific classification system for handmade and craft products. Knows that "soy wax + cotton wick + essential oil = Soy Candle" and "clay + kiln-fired + glaze = Ceramic Pottery." Contains 200+ product categories with attribute mappings. Improves Vision Agent accuracy by 34 percent by providing structured classification instead of generic object detection.

PricingEngine

Cross-references detected materials, Web Search comparables, Baidu Maps regional norms, and price elasticity data to suggest optimal pricing. Includes a "market rate confidence" score from 0 to 100. Suggests price ranges: "Similar items sell for $18 to $24. I suggest $20." Tracks price sensitivity over time: "When you dropped from $28 to $22, sales increased 300 percent."

InventoryStateMachine

Manages the DETECTED → CATALOGED → ACTIVE → RESERVED → SOLD → OUT_OF_STOCK lifecycle. Handles race conditions when two customers scan the same QR code simultaneously. Implements a 5-minute reservation window: stock is reserved when QR is scanned, released if payment fails, deducted when payment confirms. Prevents overselling and double-charging.

SalesRecapGenerator

Orchestrates Kling video generation by feeding event data (revenue, products sold, hourly velocity, best seller) into structured video scripts. Generates platform-specific formatting: vertical for Stories and TikTok, horizontal for X/Twitter, square for Instagram Feed. Includes AI-generated captions with relevant hashtags and emojis adapted to platform tone.

External API Integrations

  • ExchangeRate API: Real-time currency conversion for multi-currency stores. Updates every 24 hours. Manual refresh available. Vendor sets base currency; customer sees converted price if auto-conversion enabled.
  • IndexedDB (browser): Offline transaction queuing when markets have no wifi. Transactions stored locally, processed when connection restored. Auto-sync triggered on reconnection. Failed syncs retried every 5 minutes.
  • Shopify / Square / WooCommerce APIs: Bi-directional inventory sync for vendors with existing online stores. Webhooks receive inventory updates. API calls push PopUp sales data. Conflict resolution: external system data takes priority.
  • Social Media APIs: Auto-posting to Instagram, TikTok, X/Twitter via OAuth connections. Trigger events: store goes live, first sale, event ended. AI-generated captions with platform-specific tone.

Why PopUp

PopUp is not a POS system. It is not a website builder. It is not a marketplace. It is a one-person company operating system 也能完整运营一家公司" (One person can run an entire company).

The creator economy is $250 billion. Seventy-six percent of small businesses use AI, but only 14 percent are fully integrated. PopUp bridges that gap by making AI the invisible infrastructure behind every transaction, every marketing post, and every business decision.

The demo is visceral: one photo, 60 seconds, a working store. The business model is clear: free for 3 events, $9.99 per month for makers, $29.99 per month for merchant teams. The viral loop is built-in: every Sales Recap Video is a TikTok-ready advertisement for PopUp itself.

I built this in 47 conversations with MeDo. No traditional coding. Just describing what makers need, and watching AI employees assemble the solution.

Built With

Share this project:

Updates