Joyce

PORTFOLIO

Back to Projects
Live2026

Mi Armoire

View CodeLive Demo

ABOUT THIS PROJECT

AI styling agent that builds complete outfits with FLUX-generated visuals. Pick your occasion, choose how many looks — done.

Built with LangGraph for a four-node agentic pipeline, Groq LLaMA 3.3 70B for styling intelligence, and FLUX.1-schnell via HuggingFace Inference Router for image generation. Streamlit UI with a luxury SOLÈNE-inspired aesthetic — Cormorant Garamond, cream and gold palette, SVG geometric hero. Deployed live on Streamlit Cloud.

WHAT I LEARNED

LangGraph's node-based architecture makes multi-step AI pipelines genuinely maintainable — each node has one job and hands off cleanly

HuggingFace's Inference Router replaced direct SDXL calls after deprecation — always pin your model IDs or your image generation silently breaks

FLUX.1-schnell produces fashion-quality images when you write body-aware, specific prompts — vague prompts get vague results

Streamlit's HTML sanitiser blocks inline SVG — components.html() is the workaround for custom hero sections

Groq's speed makes a real difference in agentic loops where you're calling the model multiple times per user request

TECH STACK

LangGraphGroqFLUX.1-schnellHuggingFaceStreamlitPython

LIVE PREVIEW

Having trouble viewing? Open in new tab

Back to all projects