NoviSpace

NoviSpace is a multimodal AI design agent that gives everyone an 'Architect's Eye'—using Gemini Live to see, measure, and annotate your real-world space to ensure every purchase is a perfect fit.

Comment

Inspiration

As an architect with four years of experience, I have seen a recurring problem in the industry. While professionals can effortlessly visualize a 3D space from a 2D plan, most homeowners cannot. This visualization gap creates significant "size anxiety" during the online shopping process. Users often worry that furniture will not fit their layout or match their aesthetic, leading to high e-commerce return rates and wasted time.

I founded NoviSpace to bridge this gap. We saw an opportunity to use multimodal AI to give every homeowner an "Architect’s Eye." We wanted to move beyond static chatbots and create a living, breathing design partner that could see, measure, and validate a space in real-time.

What it does

NoviSpace is a size-aware interior design agent powered by the Gemini Live API that performs end-to-end spatial consultations.

Input: Live video feed of a room, natural voice interaction, budget constraints, and a "Design DNA" preference quiz.

Output: A structured Architectural Design Report containing:

  • Spatial Validation: Real-time measurement of walls and furniture clearances.
  • Live Annotations: Digital guidelines overlaid on the room to show functional flow and "Safe Zones" for furniture placement.
  • Design DNA Profile: A curated aesthetic summary based on interactive user swiping.
  • Budget-Aware Recommendations: Furniture suggestions cross-referenced with real-time spending limits.
  • Functional Brief: A professional-grade document for designers or retailers that includes precise dimensions and suggested layouts.

How we built it

Split-Brain Architecture (Gemini Live + Gemini 1.5 Pro):

  1. The Interaction Layer (Gemini Live API): Manages the Nora persona. This layer handles low-latency voice, manages interruptions, and provides warm, collaborative feedback. It focuses entirely on the conversational experience.
  2. The Reasoning Layer (Gemini 1.5 Pro): Monitors the live transcript and video metadata in the background. When it detects a spatial query, it triggers the measurement tools and calculates room geometry to ensure adequate circulation paths and ergonomic clearances.
  3. The Multimodal Bridge: Uses Vertex AI to process video frames. It identifies physical constraints like radiators, windows, or door frames and injects this data back into Nora’s context so she can discuss them naturally.

Infrastructure:

  • Google Cloud Run: Hosts our containerized agentic backend for high scalability and low latency.
  • Cloud Firestore: Manages real-time state for the Design DNA profiles and budget tracking.
  • Terraform: We used Infrastructure as Code to automate the entire deployment, ensuring a repeatable and production-ready environment.

Challenges we ran into

  1. Spatial Reasoning in 2D: Translating a 2D camera feed into 3D architectural advice is a complex technical hurdle. We solved this by using Gemini’s multimodal capabilities to identify reference objects and then applying architectural standards to determine realistic clearances.
  2. Server-Side Stability: We encountered known WebSocket 1011 errors when using native function calling in early preview models. This led us to develop our Split-Brain approach, which separates the voice stream from the logic calls. This made the agent robust and crash-resistant.
  3. Defining the "Architect's Voice": Making the AI sound like a professional partner rather than a calculator required extensive prompt engineering. We focused on Nora's vocabulary, ensuring she uses terms like "visual weight" and "circulation paths" while remaining approachable and warm.

Accomplishments that we're proud of

We are incredibly proud of the seamless synchronization between Nora’s verbal feedback and the visual annotations on the screen. Achieving low-latency responses where Nora can "see" a cluttered corner and immediately suggest a better-proportioned bookshelf felt like a true "Aha!" moment. We are also proud of our 100% automated deployment pipeline, which proves that NoviSpace is built on a professional-grade, scalable foundation.

What we learned

We learned that the future of spatial commerce is conversational. Moving from a text-box AI to a live-video AI completely changes the way users trust technology. People are more likely to follow architectural advice when they can see the AI "measuring" their actual home. We also discovered that "size-awareness" is not just a feature; it is the single most important utility for reducing friction in the home design and e-commerce industry.

What's next for NoviSpace

Our next step is to integrate high-fidelity AR overlays so users can "drop" the suggested furniture into their room after the consultation. We also plan to expand our Design DNA engine to include more niche architectural styles. Ultimately, we want NoviSpace to be the go-to platform for anyone moving into a new home, turning the complex process of furnishing a space into a simple, confident conversation.

Built With

  • artifact-registry
  • express.js
  • gemini-2.0-flash-live-api
  • google-cloud-run
  • lucide-icons
  • next.js-14
  • node.js
  • secret-manager
  • shadcn/ui
  • tailwind-css
  • terraform
  • typescript
  • websocket
Share this project:

Updates