MVP Architecture

ZARA
AI STYLIST

Conceptual System Architecture — High Level
Version 1.0
5 Milestones
End-to-End Flow
User Layer
AI / Intelligence Layer
Data / API Layer
Output / Generation Layer
User Layer
Entry Point

Onboarding

User lands on platform, uploads a reference photo of themselves for virtual try-on calibration.

Photo Upload Profile Store
Input

Style Query

User describes their need: occasion, dress code, color/fabric preferences, or a vibe in natural language.

Free-text Guided Filters
Action

Try-On Input

Drag & drop product images onto the try-on canvas. Select items from the results grid.

Drag & Drop Multi-select
Output

Save & Revisit

Bookmark outfits, re-visit before the event. Rate confidence 1–5. Share or purchase directly.

Saved Outfits Confidence Score
Query passed to intelligence layer
AI / Intelligence
NLP

Query Understanding

Parses free-text input. Extracts intent signals: formality, color palette, body preference, occasion type.

LLM Entity Extraction Embeddings
Matching

Semantic Search

Maps query intent to Zara product catalogue using vector similarity. Ranks by relevance and style coherence.

Vector DB Re-ranking
Curation

Outfit Assembly

Groups matched items into coherent outfits. Ensures visual harmony across top, bottom, and accessories.

LLM Style Rules
Generation

Virtual Try-On

Composites user photo + selected garments using generative model. Outputs styled portrait + optional motion clip.

Pixia / Diffusion Image-to-Video
Structured queries sent to data sources
Data / API
Catalogue

ITX-REST API

Inditex/Zara's product API. Returns live inventory: SKU, price, sizes, product images, category taxonomy.

REST API Live Inventory
Indexing

Product Vector Store

Pre-embedded Zara catalogue. Enables fast semantic retrieval without hitting the API for every query.

Pinecone / pgvector Batch Sync
Storage

User Data Store

Stores user photo, profile, saved outfits, confidence ratings, and event history. Enables revisit flows.

DB Object Store
Images

Product Image CDN

High-res Pixia-linked product imagery served for the grid results page and as input to the try-on model.

CDN Pixia Assets
Results rendered to user
Output / UI
Results

Product Grid

Responsive grid of matched garments with Pixia images, price, and size. Filterable, sortable, saveable.

Grid UI Filter Panel
Try-On

Styled Portrait

Generated image of the user wearing selected outfit. Displayed alongside product links for direct purchase.

Image Output Purchase CTA
Motion

Scene Clip

3–5 sec video of user walking in outfit with indoor/outdoor lighting variation. Enhances confidence signal.

Video Gen Lighting FX
Loop

Outfit Board

Saved looks accessible before an event. Drives revisit frequency and repeat usage signals for success metrics.

Lookbook Notifications
Delivery Milestones
01
Onboarding
Photo upload flow working. User stored, ready for try-on.
02
Search + API
Query input connected to ITX-REST. Live product data returned.
03
Grid Results
Product grid with real data and Pixia images rendered.
04
Virtual Try-On
Drag & drop → Pixia-generated output image with user + outfit.
05
Demo Ready
Full E2E flow ready. Presentation at 4PM.
Success Metrics
🛍️
Purchase
Conversion
📈
Platform
Traffic
⏱️
Try-On
Time Spent
🔖
Saved Outfits
/ User
🔁
Revisit
Frequency
Confidence
Rating 1–5
📅
Repeat Event
Usage
ZARA AI STYLIST — MVP ARCHITECTURE v1.0 Powered by: ITX-REST API · Pixia · LLM · Generative Try-On 5 Milestones → End-to-End Demo