Inspiration
Building the Missing Middle shows how housing feasibility depends on zoning, financing, and cost structure — small constraints can determine whether a project works or fails.
We were inspired by the idea that innovation in housing begins with better decision-making. Our goal was to make spatial and renovation planning more accessible by turning complex development thinking into an interactive, AI-driven experience.
⸻
What it does
Homekey is a photo-based renovation and densification assistant.
Users upload a room or property image, and the system: • Analyzes spatial layout and constraints • Suggests renovation or conversion strategies • Recommends furniture and functional layouts • Provides direct sourcing links • Enables iterative refinement through conversation
It transforms static property information into actionable planning — reducing friction between idea and execution.
⸻
How we built it
We built Homekey as a rapid prototype during the hackathon using an AI-assisted development workflow. • Cursor served as our primary development environment, enabling fast iteration through AI-assisted code generation and structured refactoring. • Codex-based tooling supported precise logic edits and modular implementation. • Google’s Gemini API powered multimodal reasoning, allowing us to combine image understanding with structured conversational intelligence in a single pipeline. • A lightweight web frontend handles image uploads, conversational state management, and dynamic rendering of recommendations and sourcing links.
Gemini’s multimodal capabilities were central to our architecture — enabling us to move beyond text-only reasoning and ground recommendations in visual spatial context.
Our system connects:
image analysis → spatial reasoning → conversational refinement → product linking
This allowed us to compress weeks of development into a working end-to-end prototype within hackathon constraints.
⸻
Challenges we ran into
Translating image understanding into actionable, spatially realistic design decisions was significantly harder than generating aesthetic suggestions. Ensuring scale, layout coherence, and constraint awareness required careful prompt and logic design.
Managing multi-turn conversational state was another challenge. Preferences around budget, style, and layout needed to persist without drift across refinement cycles.
Finally, integrating multimodal reasoning with real product linking required balancing creativity with feasibility — outputs needed to be inspiring yet grounded.
⸻
Accomplishments that we’re proud of • Successfully integrating multimodal image reasoning with conversational refinement • Delivering contextual, shoppable furniture recommendations rather than abstract design advice • Building a cohesive end-to-end system within tight time constraints • Creating a user experience that feels iterative and intelligent, not static
⸻
What we learned
Innovation is not about adding more generation — it’s about reducing friction in decision-making. When spatial understanding, reasoning, and sourcing are unified in one loop, the user moves from inspiration to execution much faster.
We also learned that multimodal reasoning combined with structured constraints produces significantly more reliable outputs than open-ended generation alone.
⸻
What’s next for Homekey • Adding zoning-aware reasoning for small-scale densification feasibility • Introducing budget optimization and cost-estimation modeling • Improving spatial grounding through depth-aware enhancements • Expanding product integrations for dynamic sourcing • Personalizing recommendations through persistent user preference modeling
Log in or sign up for Devpost to join the conversation.