Anthropic Integration¶

Sartiq integrates Anthropic's Claude API for image captioning, vision analysis, and LLM-powered outfit generation.

Overview¶

Property	Value
Provider	Anthropic
Primary Models	Claude Sonnet 4, Claude 3.5 Sonnet
Features Used	Vision, Structured Output (JSON), Text Generation
SDK	`anthropic` Python SDK v0.40+

Use Cases¶

1. Image Captioning¶

Claude's vision capabilities analyze product images to generate detailed technical descriptions for ML training and catalog generation.

flowchart LR
    IMG[Product Image] --> PREP[Image Preparation]
    PREP --> B64[Base64 Encoding]
    B64 --> API[Claude Vision API]
    API --> JSON[JSON Response]
    JSON --> CAP[Caption Output]

Supported Caption Types:

Type	Description	Use Case
`GARMENT`	Clothing construction, materials, patterns	ML training data
`PRODUCT`	Functional features, hardware, materials	Product catalog
`PERSON`	Physical characteristics, demographics	Subject casting

2. Outfit Generation¶

Claude analyzes products and generates styling recommendations through the LLM interaction system, returning structured JSON with outfit compositions.

3. Vision Analysis¶

Multi-image analysis for product understanding, style matching, and quality assessment.

Architecture¶

flowchart TB
    subgraph Backend["Backend API"]
        CS[CaptionsService]
        LLM[LLM Interaction Handler]
        AIC[AI Core Client]
    end

    subgraph Providers["Provider Layer"]
        AP[AnthropicProvider]
        LB[LiteLLM Backend]
    end

    subgraph External["Anthropic API"]
        CLAUDE[Claude API]
    end

    CS --> AP
    LLM --> AIC
    AIC --> AP
    AP --> LB
    LB --> CLAUDE

Models¶

Alias	Model ID	Use Case
`claude`	`claude-sonnet-4-20250514`	Default, captioning
`sonnet`	`claude-sonnet-4-20250514`	General tasks
`claude-3-5-sonnet`	`claude-3-5-sonnet-20241022`	Outfit generation
`haiku`	`claude-3-haiku-20240307`	Lightweight operations
`opus`	`claude-opus-4-20250514`	Advanced analysis

Key Features¶

Structured JSON Output¶

All integrations request JSON responses with defined schemas:

# Captioning response schema
{
    "technical_description": "...",
    "short_description": "..."
}

# Outfit response schema
{
    "core_product": {...},
    "looks": [...]
}

Vision Capabilities¶

Base64 image encoding with automatic format detection
Size optimization (max 5MB)
Support for JPEG, PNG, WebP, GIF
Multi-image analysis

Error Handling¶

Retry logic with 3 attempts on failure
Graceful JSON parsing (handles markdown code blocks)
Rate limit guidance with model fallback suggestions

Documentation¶

Section	Description
Setup	Configuration and API keys
Prompting	System prompts and best practices
Examples	Code examples and workflows