fashion-stylist-agent

Fashion Stylist Agent is an agentic AI system that recommends personalised outfits based on occasion, gender, weather, and time of day. It combines LangGraph orchestration, multimodal RAG retrieval, and generative image synthesis. The project is built as a portfolio project demonstrating production-oriented AI engineering in a fashion context.

The system uses a LangGraph agent with five tools: a Context Builder that creates a style brief from raw inputs, a RAG Retriever using FashionCLIP and ChromaDB for semantic item search, an Outfit Composer that selects a structured JSON outfit, an Image Generator using DALL-E 3 or SDXL for outfit visualisation, and a Stylist Narrator that provides an editorial outfit explanation. The final output is an image, outfit pieces, and stylist narration.

Key design choices include using LangGraph over bare LangChain for a stateful graph with conditional retry logic, FashionCLIP over generic CLIP because it was trained on 1M+ fashion items and offers 57% better text-to-image retrieval, ChromaDB with metadata filtering for gender-aware retrieval, and dual-mode generation with free local SDXL during development and DALL-E 3 for the final deployed demo. Recommendations come only from items ingested from the Polyvore-U dataset, and the outfit composer relies on the LLM returning valid JSON, with retries up to twice before stopping with an error.