Our Technology

8 AI Engines Working Together

Closetary isn't a simple app with a ChatGPT wrapper. It's a purpose-built AI platform with specialised engines for styling, sustainability, and visual intelligence.

Architecture

How It All Fits Together

Backend

FastAPI (Python)

13 feature modules

Mobile

Kotlin Multiplatform

iOS + Android

Database

PostgreSQL + pgvector

Vector similarity search

AI

Google Gemini

Pluggable LLM provider

Outfit Generation Pipeline

1

Anchor Selection

Pick a hero item

2

Vector Retrieval

pgvector cosine similarity

3

Rule Filtering

Weather · Occasion · Fit

4

LLM Generation

Styling advice via Gemini

AI Engines

Purpose-Built Intelligence

Fusion Engine

Outfit Generation

Our core outfit generation pipeline combines four stages: anchor item selection, vector similarity retrieval via pgvector, rule-based filtering (weather, occasion, dress code), and LLM-powered styling advice.

pgvector cosine similarityRule engineGoogle Gemini LLMPluggable provider architecture

CLIP Embeddings

Visual AI

Generates vector representations of garment images using CLIP (Contrastive Language-Image Pre-training). Enables visual similarity search, cross-item matching, and authenticity confidence scoring.

Vector embeddingsCosine similarityVisual searchBrand verification

Carbon Calculator

Sustainability

Lifecycle Assessment (LCA) engine calculating per-item carbon footprints. Aligned with PAS 2050, using UK DEFRA emission factors, WRAP UK data, and Higg Materials Sustainability Index.

PAS 2050 alignedDEFRA emission factorsHigg MSI dataMaterial composition parsing

Stylist Orchestrator

AI Coordination

High-level orchestrator combining the Fusion Engine, Carbon Calculator, and user preferences to generate context-aware, sustainability-conscious outfit recommendations.

Multi-engine coordinationUser preference learningWardrobe gap analysisIdle item detection

LLM Framework

Foundation

Abstracted language model provider with factory pattern. Currently using Google Gemini, but designed to be swappable between providers without changing business logic.

Provider abstractionFactory patternGemini integrationMock provider for testing

Mobile ML

On-Device

On-device machine learning capabilities for instant garment recognition and categorisation directly on the user's phone, without requiring a network connection.

On-device inferenceOffline supportLow-latency recognitionPrivacy-preserving

Sustainability Is Engineered In

Every AI recommendation considers environmental impact. Our Carbon Calculator scores each garment using research from UK DEFRA, WRAP UK, and the Higg Materials Sustainability Index — not generic estimates, but material-specific lifecycle assessments.

PAS 2050

Carbon methodology

DEFRA

UK emission factors

Higg MSI

Material scoring

WRAP UK

Lifecycle research

Experience the Technology

Join our waitlist to be first to try Closetary's AI-powered platform.

Join the Waitlist