Déjà View | Devpost

Inspiration

When people are asked who they are, they often reach for a name, a title, a label. Identity isn’t made of facts, but the colors you gravitate toward, the corners you return to, the objects you keep the world you build around you.

We built Déjà View around that idea: if your room is part of your identity, then the things you discover online shouldn’t stay trapped in a “saved” folder. They should step into your space, be felt, noticed, and chosen.

What it does:

Déjà View turns online inspiration into a living virtual room.

Scan your space Record a 45–90s video (or LiDAR scan) → we generate a 3D room.

Collect inspiration: Save Pinterest posts to a folder → Déjà View pulls them in.

Bring it into your room: We identify the item, generate a realistic 2D render, convert it into a 3D object, and place it naturally in your space.

Decide with context: Tap the object to favourite, add to cart, or dislike.

Shop instantly: We generate tags and surface real matching products on Shopify.

Resonance Test: Items drift daily for 3 days, if you never interact, they disappear.

How we built it

We built a multi-step pipeline that connects discovery → creation → placement → purchase:

Next.js + TypeScript for the web app

Tailwind + custom tokens + Framer + GSAP for a clean, animated UX

Three.js for the 3D room viewer and object rendering

Clerk for authentication

MongoDB Atlas for app state + metadata

Cloudflare R2 for storing generated assets (2D images + 3D models)

Pinterest API for ingestion → Gemini API to identify the object + generate tags/description

Nano Banana Pro to generate a realistic 2D product render

Trellis to convert the 2D render into a 3D object

Gemini again to infer a natural placement in the room

Shopify Storefront API to display similar real products and enable checkout

Challenges we ran into

Video/LiDAR → 3D room conversion: turning casual room capture into a usable interior space was harder than expected.

Accurate 3D model generation from Pinterest images: chaining Pinterest → AI prompt → 2D render → 3D conversion meant every step needed guardrails and fallbacks.

Placement that “makes sense”: it’s not enough to place an object somewhere vali, it needs to feel natural in context (floor vs surface, near walls, not clipping, not floating).

Reliability across many APIs: this project is basically a relay race; if one handoff fails, everything fails, so we built the pipeline to be resilient.

Accomplishments that we're proud of

We got the full pipeline working: Pinterest → Gemini → Nano Banana (2D) → Trellis (3D) → placed in-room.

We built a polished, modern experience with micro-interactions and motion that makes the product feel production-ready.

We connected the identity experience to a real outcome through Shopify purchase flow.

What we learned

Building “multi-step, multi-API” systems is less about any single model, it’s about designing clean interfaces between steps, adding fallbacks, and pushing through integration friction.

Identity changes over time. We learned that good design helps people make choices on purpose, so what they keep actually reflects who they are.