Arcana (Interior Design AI Agent)

Inspiration

We started Arcana to solve a problem we’ve all experienced, how hard it is to visualize a space before buying furniture. Professional design costs hundreds per hour, DIY tools are slow and clunky, and most AI tools only generate pretty pictures without real-world accuracy. We wanted to build something smarter an AI system that truly understands geometry, style, and space and makes interior design as intuitive as asking a question.

What it does

Arcana is an AI-powered interior design platform that transforms any room photo or sketch into a photorealistic, geometrically accurate layout with real, purchasable furniture. Using a multi-agent Claude architecture, Arcana’s specialized agents handle style analysis, spatial planning, budget management, and product sourcing, while ControlNet generates lifelike visuals and our geometric CAD engine ensures everything fits perfectly.

How we built it

• Claude Opus + Sonnet 4: Orchestrator and worker agents for design reasoning, style extraction, and product recommendations.
• ControlNet + Stable Diffusion: Image-to-design generation from sketches and photos.
• Custom CAD Engine: Parametric geometry modeling for layout optimization and clearance checks.
• FastAPI + React + Tailwind: Full-stack integration for a seamless, real-time user experience. ## Challenges we ran into
• Coordinating multiple Claude agents with structured outputs and token-efficient communication.
• Balancing photorealism with geometric accuracy — ensuring designs looked beautiful and fit real dimensions.
• Fine-tuning ControlNet to produce consistent lighting and perspective across different room inputs.
• Debugging RAG retrieval pipelines to return contextually relevant furniture items under time pressure. ## Accomplishments that we're proud of
• Built a fully functional multi-agent design intelligence system within 36 hours.
• Integrated Claude tool use + contextual RAG
• Created a live geometric CAD visualization for real-time layout validation ## What we learned
• Multi-agent coordination with Claude massively improves reasoning quality and modularity.
• Contextual embeddings are essential — retrieval improves dramatically when chunks include design context, not just text.
• CAD and AI image generation complement each other: geometry validates AI creativity, and visuals inspire layout refinement ## What's next for Arcana (Interior Design AI Agent)
• Expand CAD integration into full 3D generative geometry modeling
• Real-time collaboration between human designers and AI agents using LiveKit and voice interaction.
• Extend contextual RAG to support millions of furniture items with automatic catalog ingestion.

Built With

• claud
• fastapi
• javascript
• python
• rag
• react
• tailwindcss