Déjà View

From inspiration to identity.
Turn your Pinterest saves into 3D objects in your room — keep what you touch, let the rest fade.

demo.mp4

Landing Page

3D Room View

The Pipeline

See how Déjà View transforms your Pinterest inspiration into interactive 3D objects:

Pinterest	2D Render	3D Model
Source inspiration	AI-extracted product	Generated 3D model

How It Works

1. Save: Collect inspiration on a public Pinterest board
2. Import: Paste your board URL — we analyze each pin and extract the main item
3. Generate: AI creates a realistic 2D render, then converts it to a 3D .glb model
4. Place: Objects appear naturally in your virtual room
5. Interact: Click items to inspect, adjust placement, or shop similar products
6. Curate: What you touch stays; what you ignore fades — your space reflects you

Quick Start

Prerequisites

• Node.js 18+
• Python 3.10+
• API keys: Clerk, Gemini, Replicate, Cloudflare R2, Google CSE

Frontend (Next.js)

cd deja-view
npm install
cp env.example .env.local
# Fill in your API keys
npm run dev

App runs at http://localhost:3000

Backend (FastAPI)

cd pinterest-extraction
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
cp env.example .env
# Fill in your API keys
uvicorn app.main:app --reload --port 8000

API docs at http://localhost:8000/docs

Architecture

flowchart LR
  U[User] --> W[Next.js App]

  subgraph Frontend["Next.js App Router"]
    W --> A1[/api/pinterest-board/]
    W --> A2[/api/items/]
    W --> A3[/api/get-placement/]
    W --> A4[/api/match-cache/]
  end

  subgraph Backend["FastAPI :8000"]
    P1[/v1/analyze-job/]
    P2[/v1/extract-item-3d/]
  end

  A1 -->|start + poll| P1
  A1 -->|start + poll| P2

  P1 -->|LLM analysis| LLM[OpenAI]
  P2 -->|item extraction| G[Gemini]
  P2 -->|3D generation| R[Replicate Trellis]
  P2 -->|store assets| R2[Cloudflare R2]

  A2 --> DB[(MongoDB)]
  A3 -->|placement inference| GV[Gemini Vision]

Loading

Tech Stack

Frontend Next.js 14 & React Three.js for 3D rendering Clerk for authentication Framer Motion & GSAP for animations

Backend FastAPI with async job polling Gemini for item extraction & placement Replicate Trellis for 3D .glb generation MongoDB for data persistence Cloudflare R2 for asset storage

Inspiration

University students express identity through their space, but buying furniture from flat images leads to regret and costly returns. Inspiration stays trapped in static Pinterest boards — disconnected from the real constraints of your room.

Déjà View bridges that gap: discover → visualize in 3D → decide with context → shop what truly belongs.

Team

Name	Role
David Liu	Web app, UI/UX, product, video/LiDAR → 3D room model
Julian Laxman	Pinterest integration, Pinterest → 2D → 3D pipeline, backend integration
Alan Xue	Shopify integration, UI/UX, Web app, Google PSE API -> MongoDB pipeline integration
Charles Bae	3D room modeling & object placement logic, Three.js integration, system components integration

---

Built at UofTHacks 13