Right: Designer’s first interpretation → Earthy, Sculptural and Serene
How I Framed the Problem
Problem Statement
In home renovation, clients often describe their dream spaces using abstract terms like “calm” or “natural.” Designers must decode these emotional inputs into clear decisions around layout, materials, and budget. This gap causes frequent misalignment, delays, and unnecessary revisions.
Clients can tell you what they don’t want, but rarely what they actually mean. "
— Senior Interior Designer, UK
“
Client's Expectation Vs. Designer's Interpretation
Left: What the client said → “Calm", "Modern" and "Earthy"
KEY INSIGHT
It was a translation gap between client intent and execution. Instead of automating decisions, I focused on building a system that interprets ambiguity without taking away designer control.
This wasn’t a tooling issue.
KEY QUESTION
How might we...
...translate emotional client input into visual clarity without limiting a designer’s control?
Left: What the client said → “Calm", "Modern" and "Earthy"
Right: What Stellar produced → Conceptual render using AI tags
Turning emotional input into structured design output
In our MVP, ‘clients’ referred to homeowners working with interior design studios to renovate their spaces.
How We Built It
Overview
COMPANY
Stellar
PLATFORM
Web
TEAM
Myself (UX), 1 PM, 1 ML Engineer, 2 Developers, Board of Industry Experts
USERS
5 design studios, 25 designers, 30+ live projects
What We Built
Stellar is an AI-powered assist tool for home renovation studios. It helps designers translate emotionally vague client input (Example: “calm", "earthy” etc.) into structured visual directions and AI-generated renders.
Pilot
The 3-month MVP beta was piloted across five design studios, including our advisory board of industry experts, who also served as both primary users and co-creators.
STELLAR
Decoding vague client input using AI to boost
design concept accuracy by 30%
IN THE HOME RENOVATION INDUSTRY
Project at a Glance
Problem:
Clients often describe their style preferences using emotional or abstract words, which makes it hard for designers to get the clarity they need. This leads to misalignment, delays, and extra revisions.
My Role:
Led end-to-end UX, from research to Figma prototyping, and collaborated with PM, ML engineers, and industry experts.
Validated Impact:
All metrics were derived from actual MVP pilot sessions across 30+ active projects over a 3 month period.
Automation isn’t the goal. What mattered was giving designers a confident, aligned starting point that saved time without sacrificing creativity. "
— Partner Studio Lead
“
What We Needed Next
To resolve the disconnect, we needed an AI that could partner with the designer by balancing creative input with intelligent automation. This shaped how we approached training the AI, with a focus on interpretability, emotional resonance, and preserving designer control.
Visualization challenges
Material Sourcing Delays
repeated revision cycles
Video Credit: Author
Excerpt from Interview with the Postgraduate Program Lead, School of Architecture, Oxford Brookes University
How I Mapped the Research
Research Scope
I led 25 interviews with interior designers and architects across Dubai, Egypt, the UK, the US, and India. The goal was to understand friction across the end-to-end client–designer workflow.
Workflow Mapping
I synthesized findings into a simplified journey map to highlight friction across each stage of a residential project. The Planning and Execution phases consistently surfaced as the most problematic, often triggered by vague input and scope creep.:
ACUTE pAIN pOINTS ACROSS HIGHLIGHTED STAGES
.png)
Key Patterns Identified
During initial consultations, designers experienced recurring breakdowns in alignment. I grouped these into three clear patterns:
Unstructured inputs
communication gaps
budget ambiguity
.png)
Vendor coordination
late-stage changes
scope overruns
quality assurance
.jpg)
Read Between the Lines
Client input use emotional language, leaving designers to interpret feelings into detail.
Mood boards and reference images helped but often missed nuances in tone, texture, or context.
Band-Aid Fixes
Most projects required 3–5 rounds of revisions, delaying timelines by 2–4 weeks.
Rework Loops
I spend more time clarifying what the client actually wants than on design itself. "
— Interior Designer, US
“
Tagging Dataset
To fine-tune Stellar’s tagging engine, I curated a dataset of architectural renders spanning styles like Japandi and maximalist to trigger honest reactions from both professionals and non-professionals.
How We Trained the AI
Session Structure
To teach the AI how to bridge designer vocabulary with client intent, I led structured tagging sessions with two distinct user groups. Each participant reacted to the curated set of architectural renders by tagging what they saw or felt.
Non-Designers
Homeowners and lifestyle-focused users without formal design training
Tags used:
Designers
Professional architects and interior designers familiar with technical terminology
Tags used:
What stood out
Non-designers described how the spaces made them feel using sensory and emotional cues linked to lifestyle and mood.
Designers focused on specific design attributes such as materiality, light quality, and spatial layout using formal terminology.
What the AI Learned
Each image received both emotional and technical tags. This dual-tagging method trained the AI to recognize patterns across vocabularies and align them to real-world outputs.
Video Credit: Author
Tagging Exercise With Designer
Non-designer Tags | Designer Tags | Combined AI Output |
|---|---|---|
Diffused lighting | Diffused lighting | Ambient lighting with sheer curtains |
Calm | Matte finish | Warm matte walls with minimal texture |
Open, breathable | Modular layout | Flexible layout with open seating zones |
Why It Mattered
This approach ensured the AI was:
• Transparent and explainable
• Emotionally resonant for clients
• Practical and editable for designers
It was the first time I felt like an AI model understood what I meant by natural. Not just slapping plants on everything."
— MVP Client
“
MVP Strategy
The Guiding Principle
We structured Stellar’s roadmap around one core question:
Will this reduce misalignment within the first 10 minutes of a client–designer interaction?
Prioritization Approach
I led a 2×2 impact-effort working session with our PM, ML engineer, and two partner studios. Instead of grouping features by type, we focused on those that most directly addressed vague client input, the root cause of early misalignment.
MVP Phases We Prioritized
Features We Deferred
