AI Fashion Research Assistant — 2025 Prototype

Historical project: This was a May 2025 Perplexity Hackathon prototype. It is not the current Caeliai service.

This project explored how AI-assisted web research could help people identify fashion items, understand product context, compare market information, and investigate brand materials and manufacturing signals.

The work began with a narrow focus on AI and fashion. The research questions, structured-output experiments, and product-discovery lessons from this prototype later helped inform Caeliai, which has since expanded into broader AI-commerce research for ecommerce brands.

Current Caeliai research: caeliai.com

What the prototype did

Users interacted with a chat interface to:

• Identify fashion items from text descriptions
• Find retail and resale information
• Retrieve design and runway context
• Research brand materials, manufacturing locations, and certifications
• Return structured answers powered by the Perplexity Sonar API

How it was built

• Next.js 14 and TypeScript
• React
• Supabase Auth and PostgreSQL
• Supabase Edge Functions
• Perplexity Sonar Pro API
• JSON Schema structured outputs
• Vercel deployment

What I learned

This prototype demonstrated the importance of:

• Reliable structured output from language models
• Clear source constraints and domain filtering
• Iterative prompt design
• Connecting AI-generated product research to verifiable web evidence
• Distinguishing an experimental prototype from a repeatable measurement system

How the work evolved

After the hackathon, the focus moved beyond individual fashion-item research. Caeliai now studies how ChatGPT and Gemini recommend ecommerce products, which brands appear, and where AI-generated purchase links send buyers.

The original fashion research remains part of the project history, while the current work is published separately at caeliai.com/research.

Built With

• javascript
• json-schema
• next.js
• perplexity
• postgresql
• react
• sonar-api
• supabase
• typescript
• vercel