With preliminary product vision and research, Dwellci AI is struggling with a functional MVP to validate concepts and test out user feedback.

The Solution

As the only designer on the team, I built key flows and AI features to support project and study creation, as well as AI model generation and optimization. The MVP received 4.88/5 in preliminary user testing.

Dwellci AI

An AI-led Transformation of “How Architects Might Work?”

Architects are known to go through complex development stages, where they gather client and site requirements, brainstorm multiple ideas, explore feasibility, and hop onto different softwares like AutoCAD, Revit, Rhino to create polished models and rendering.

AI is coming in to change how they work. Starting with Dwellci AI, aiming to empower the early conception phase with AI — making the process faster and freer.

The Problem

User Research Findings

Limited Exploration

While creating design details are time consuming, architects have to think about site constraints and client requirements at the same time, making it hard to make free explorations.

Complicated Workflow

To create 2D models, 3D models, and track data sheets hosted by different softwares, they have to bounce between multiple softwares and files simultaneously. What’s more, update in one file requires simultaneous updates of other files too.

Repetitive Creation

Drawings lose details when they export to other platforms. This often requires them to recreate some of the details when exporting to another software. Sometimes, they also have to create simplified version of models for client representation.

Interview notes & Affinity Mapping. KZ.

Findings on AI Capacities

AI can generate drawings and 3D models.
Users can chat with AI to make design modifications.
AI changes can be integrated into model real-time updates.

AI capacity Concepts. KZ.

How Might We ensure AI generated models are usable and understandable?

AI Generation Flow Comparison

AI needs data input first to generate models. Therefore, I created 3 flows with different types of data input and discussed with the founder about their feasibility. Considering front loading data constraints can greatly ensure model accuracy to user requirements, I decided to go with the second flow.

AI generation flows. KZ.

AI generation wireflows. KZ.

To simplify the data input process, I broke them into 4 smaller steps to guide users step by step.

AI generation wireflows 2.0. KZ.

Dwellci AI MVP Demo Clip. KZ.

AI generation flows 2.0. KZ.

Competitive Analysis on Google Delve. 2025.

Feature: efficient, reusable site creation for multiple studies.

Dwellci AI MVP Demo Clip. KZ.

Feature: guided data input and AI generation.

How Might We allow users to modify models easily?

The Current Flow

Currently, architects are jumping back and forth between different softwares hosting 2D models, 3D models, data sheets, requirements, etc. Scenarios include: updating 3D models because 2D model changed, checking codes white drawing up 2D models, updating plans while data changes…

The canvas should integrate views, data, and AI chats.

I explored the first layout of the design canvas that attempts to integrate the ideal features, and explored ways to make the interaction more intuitive and experience friendly.