Back to Projects

POD AI – Intelligent Personalization Campaign Engine

E-CommercePrint on DemandGenerative AIProduct PersonalizationShopBaseWeb App
Team Size: 3 member(s)
POD AI – Intelligent Personalization Campaign Engine

Introduction

In the Print-on-Demand (POD) industry, personalized products are the ultimate revenue drivers. A custom hoodie featuring best friends with matching hairstyles, or a ceramic mug showing a customer sitting next to their beloved dog breed with their names inscribed, consistently commands double the selling price of generic merchandise. Shoppers eagerly pay a premium when an item reflects their personal story.

However, behind every successful personalized listing lies an agonizing operational nightmare that every POD merchant knows all too well: the manual setup process.

The Merchant's Core Pain Point: Painfully Slow & Exhausting Manual Configuration

To launch a compelling personalized product, the volume of visual assets and conditional combinations a seller must configure manually is staggering.

📌 A Real-World Example: "Best Friends & Pets" Campaign

Imagine setting up a t-shirt featuring 2 best friends and 1 pet dog:

  • Character 1: 25 hairstyles × 8 hair colors = 200 hair cliparts. Add 10 body poses, 15 outfits, and 5 skin tones = hundreds of individual image files.
  • Character 2: The exact same complexity, adding another 250+ image assets.
  • Pets: Shoppers own different breeds. To satisfy customer demand, the merchant must offer at least 40 popular dog breeds (Corgi, Golden Retriever, Poodle, French Bulldog, German Shepherd...), each with 3 to 4 distinct fur color variations = over 150 pet cliparts.
  • Custom Text: Text inputs for each person's name, the pet's name, and custom commemorative quotes.

👉 In total: A single campaign easily exceeds 600 to 1,000+ individual graphic assets and layer combinations!

😫 What Merchants Endured Doing This Manually:

  1. 2 to 4 Hours per Product: The merchant had to manually slice layers in Photoshop, export hundreds of transparent PNG files, upload each file one-by-one into their store backend, and painstakingly name every single asset.
  2. A Tangled Web of Conditional Rules: Sellers had to manually map hundreds of tedious conditional rules: "If the buyer picks Short Wavy Hair, show these 8 color swatches; if they select a Golden Retriever, restrict fur colors to Golden/Cream...".
  3. Severe Risk of Costly Human Error: A single mistyped layer name or misconfigured condition caused disasters—such as a buyer ordering blonde hair but the shirt printing out with black hair, or hair floating off-center from the torso. The factory printed defective items, customers demanded refunds, and merchants took heavy financial losses.
  4. Missing Lucrative Trends: When an idea went viral on TikTok or Etsy, manual setup consumed half a day—by which time competitors had already captured thousands of sales.

How POD AI Solves This Challenge

POD AI (AI-Personalized Campaign) was built around one singular mission: Transforming a 3-hour manual ordeal into an automated 5-minute setup.

The system unites the entire journey into a continuous, automated flow:

text
[ 1. UPLOAD ARTWORK ]      [ 2. CHOOSE MODEL ]         [ 3. EDIT WITH PROMPTS ]    [ 4. PUBLISH & PRINT ]
Auto-detect characters  ──►  Details only or        ──►  Natural language layout  ──►  Live shopper preview
and options in 5 mins       Count & details (<=7)        Smart composition polish       Direct 300 DPI output

1. Upload Artwork & Automatic Character Detection

  • Merchants simply upload their master design file (PNG, JPG, JPEG with transparent background).
  • The AI scans the composition and automatically detects characters (people, pets, or objects).
  • Merchants can easily adjust character counts (Set character count) or add custom entities (Add Custom).
Slide 1
Slide 2
Slide 3

2. Flexible Personalization Models

The platform supports two distinct customizer archetypes:

  • Details only: Applies global character traits (hair styles, skin tones, pet breeds) equally across all characters of the same type.
  • Count and details: Empowers shoppers to choose how many characters appear (e.g., 1, 2, or 3 pets) AND customize individual details for each character independently (supporting up to 7 characters per design).
Personalization models: Details only vs Count and details

3. Automated Options, Swatches & Conditional Logic

  • Supports diverse UI input types: Picture Choice, Color Swatch, Select Option, Text Input, and Number Tiles.
  • Automated Conditional Logic: The engine automatically infers contextual pairing rules (e.g., selecting a specific cat breed automatically restricts the visible color swatches to matching fur patterns).
  • In-app generative AI allows sellers to synthesize new cliparts and background assets on the fly.
Slide 1
Slide 2
Slide 3

4. Breakthrough: Prompt-Based Layout Editing & Storefront Preview

  • Edit with prompt: Instead of tedious manual dragging, sellers can simply type plain-text natural language prompts (e.g., "position the characters closer together", "shift the pet to the right side"), and the AI recalibrates the layout automatically!
  • Improve Composition & Fix Characters: One-click automated visual balance optimization.
  • Preview Storefront: Interactive sandbox showing exactly how buyers will experience the customizer before pushing the campaign live.
Slide 1
Slide 2
Slide 3

5. Multi-Product Mockups & 1-Click Campaign Launch

  • Synchronizes the custom artwork across apparel, mugs, hoodies, and canvas wall art.
  • Remove background when generating preview: Intelligently strips base backgrounds during live preview rendering, ensuring artworks sit naturally on physical garment mockups.
  • Set retail pricing, add tags, and hit Publish to launch across global storefronts in seconds.
Slide 1
Slide 2
Slide 3

6. Real-Time Buyer Personalization & Automated Fulfillment

  • Shoppers visit the store, freely experiment with styles and names, and enjoy instantaneous Live Previews on realistic apparel mockups.
  • Upon checkout, the system automatically composites custom vector layers into a print-ready 300 DPI vector file, routing it directly to the manufacturing facility with zero manual designer touchpoints.

My Role & Contributions

Working in a 3-member engineering team, I developed the core functional mechanics:

  • Intelligent Swatching & Conditional Rules: Built the attribute-pairing logic connecting visual silhouettes with color swatches, freeing sellers from manually scripting hundreds of conditional rules.
  • End-to-End User Experience: Developed both the seller configuration console in the administration dashboard and the live customization widget for shoppers on storefronts.
  • Live Preview Performance: Optimized client-side layer compositing to deliver instant visual feedback on every user interaction with zero storefront page lag.
  • Print File Integrity: Built the high-resolution 300 DPI artwork synthesis pipeline, ensuring that the physical item printed at the factory matches the customer's on-screen preview with 100% fidelity.

Tangible Business Value

  • 95% Reduction in Setup Time: Slashed product creation time from 2–3 hours down to under 5 minutes, allowing merchants to capitalize on trending viral designs immediately.
  • Zero Human Error: Completely eliminated misaligned layers, incorrect hair colors, and print mismatches caused by manual data entry.
  • Surging Conversion Rates: Engaging real-time visual previews drove substantial uplifts in buyer engagement and purchase completion rates.
Live Demo