SOZOColour: Search by Seeing

What the company appears to be: SOZOColour is a self-reported AI-powered colour-search tool designed for colour-led retailers. It allows users to upload images or select reference colours, then translates visual intent into product recommendations from a merchant’s catalogue using both perceptual colour science (CIEDE2000) and natural language interpretation via GPT-5.6.

What changed: The project evolved from an exploratory prototype into a production-ready B2B2C experience during a Build Week hackathon, integrating image-based selection, deterministic matching, and intent-aware AI reasoning over a real product catalogue.

Single most important open question: Is there evidence of traction or commercial adoption beyond the Nail Deck use case? The description does not state whether SOZOColour has been deployed with other merchants or generated revenue. Without such data, it is unclear if this is an emerging product or a proof-of-concept.

The description states that SOZOColour is an AI colour-reasoning layer for colour-led commerce. It enables users to upload images or select reference colours and receive product recommendations from a merchant’s catalogue based on both objective visual similarity (using CIEDE2000) and subjective intent interpreted through GPT-5.6.

It separates two distinct user jobs:

1. Closest to what you saw — determined by perceptual colour matching.
2. Closest to what you meant — grounded in natural language input about feeling, context, or desired transformation.

The system is built with:

• Browser-based image upload and manual colour selection
• Server-side processing using CIEDE2000 for matching
• GPT-5.6 via OpenAI Responses API for intent interpretation
• A secure API boundary to protect keys
• Structured output validated by product IDs

It supports a shopper-facing experience that presents both objective matches and intent-aware suggestions, distinguishing between visual similarity and perceptual fit.

The description states that SOZOColour addresses the challenge faced by colour-led retailers where customers recognize shades visually but cannot name them effectively. It positions itself as a discovery tool that helps shoppers express otherwise unsearchable intent.

It claims to be different from general visual-search tools because it:

• Combines perceptual colour measurement with semantic interpretation
• Focuses on transforming ambiguous human language into actionable product decisions
• Operates within a controlled set of real products rather than inventing new ones

The author also notes that the tool is built for merchants, not end-users directly — suggesting a B2B2C model where retailers deploy it as part of their storefront.

The description states that SOZOColour targets colour-led retailers such as:

• Nail Deck (the first working deployment)
• Interior paint and wallpaper
• Cosmetics and hair colour
• Fashion, footwear, and accessories
• Furniture and home décor
• Stationery, textiles, ceramics, and materials

It is positioned for merchants who want to improve product discovery in categories where colour plays a central role in purchasing decisions.

The ICP appears to be:

• Retailers with large, curated product catalogues
• Businesses that rely heavily on visual appeal and subjective perception for sales
• Companies looking to enhance their digital commerce experience through AI-driven personalization

No specific customer segments or personas are named beyond the industries listed.

The description states that the initial commercial model is B2B2C, where merchants deploy SOZOColour, while shoppers use the embedded discovery experience.

Potential delivery models mentioned include:

• Commerce-platform app
• White-label widget
• Usage-based API
• Enterprise catalogue integration

There is no mention of pricing tiers, usage limits, or monetization strategies beyond deployment and embedding. No revenue data or customer contracts are provided.

The description provides technical details about how SOZOColour was implemented:

• Uses CIEDE2000 for perceptual colour matching
• Implements GPT-5.6 via OpenAI Responses API
• Employs a Node.js server with secure handling of API keys
• Includes deterministic fallbacks when AI requests fail
• Separates objective and intent-aware recommendation results
• Supports responsive layouts and structured output validation

It also mentions that Codex was used as the primary development collaborator, documenting architecture decisions and iterations.

The system is described as production-grade for Build Week but lacks evidence of long-term scalability or performance metrics.

Not evidenced.

The description does not provide any data on:

• Number of users
• Conversion rates
• Merchant adoption beyond Nail Deck
• Product usage frequency
• Customer feedback or retention

It only describes the first deployment at Nail Deck and a prototype built during Build Week. No evidence of traction, growth, or user engagement is present.

Not evidenced.

The description does not mention:

• Competitors in the visual search or colour-matching space
• Market size or competitive landscape
• Differentiation from existing tools like hex value matchers or general visual-search engines

It only asserts that SOZOColour is different from typical tools by combining perceptual science with intent-aware AI, but no comparative analysis or market positioning is offered.

1. No commercial traction: The project has not demonstrated adoption beyond Nail Deck or any measurable user engagement.
2. Unverified claims: The description makes strong claims about intent interpretation and perceptual matching without evidence of accuracy or effectiveness.
3. Limited scope: The only known deployment is Nail Deck, with no indication of expansion plans or broader market validation.
4. Dependency on AI model: Reliance on GPT-5.6 raises questions about consistency, cost, and availability if the API changes or becomes unavailable.
5. Unclear monetization path: While a B2B2C model is proposed, there’s no clarity on how merchants will pay or what value they derive.

1. What is the current status of the Nail Deck deployment? Is it live, and if so, how many users interact with it?
2. How does SOZOColour handle edge cases in colour perception (e.g., lighting conditions, screen calibration)?
3. Are there any known limitations or biases in GPT-5.6's interpretation of subjective language?
4. What are the technical constraints or bottlenecks in scaling this solution to other retailers?
5. Has the team conducted any user testing or A/B experiments to validate the effectiveness of intent-aware recommendations?
6. How is data privacy handled, especially when users upload images and provide personal descriptions?