Djarvis

Residential · Plan · Render

From sketch to dimensioned plan to bound 3D.

Djarvis is an AI-assisted studio for residential architecture. The pipeline is three deterministic stages — analyze, lay out, render — and the 3D renderer never invents geometry. Every drawing has a structured JSON sibling. Every artifact is immutable, versioned, and addressable in chat.

01

Analyze

Upload a sketch, photo, scan, or PDF. Vision extracts walls, openings, and rooms with millimeter dimensions and a confidence score. Edit before you proceed.

02

Lay out

Generate N candidate furniture arrangements over the same wall set. The 2D drawing is rendered deterministically from the JSON — never AI — so what you see is what is.

03

Render

3D follows the plan. The renderer is bound to the metrics; geometry-changing prompts are routed back to the layout. Style, material, light, decoration — those you can change freely.

R1

3D follows the plan

The renderer reads a frozen LayoutMetrics; output is verified by a vision check.

R2

Source of truth is structured

Every artifact has a JSON sibling. Pixels are derivative.

R3

Stages are a DAG

Enter at any step. Operations gate on artifact preconditions, not project status.

R4

Artifacts are immutable

“Modify” always produces a new version with a parent reference.

R5

Providers swap, not rewrite

One adapter interface — adding a model is a config row.

·

Five ways to enter the pipeline

  • Upload & Analyze

    Sketch / photo / PDF

  • Import Layout

    A finished 2D plan

  • Import Render

    An existing 3D image

  • Single Room

    Photo of one room

  • Blank

    Type metrics directly