SORTA AI · 2025
Role
Founding Designer
TEAM
Team of 6
stage
pre-seed/exploration
duration
6 months
What I did here
my focus
Feature prioritization
Narrowed 40+ ideas to a single validated feature set
MVP validation
Ran lightweight tests before any engineering investment
the outcome
One clear product direction, team aligned, shipped in 6 weeks.
The situation
reimagine the human-AI relationship
I've been thinking a lot about the future of human-AI relationships.
Right now, most AI is command-driven. You prompt. It responds. You manage the output.
That's still cognitive work — just outsourced differently.
The question I kept coming back to:
What if instead of adding another interface to manage, AI could learn how you live and quietly reduce the decisions you need to make?
Not an agent that executes.
Not a copilot that assists.
A companion that adapts.
That's the design hypothesis behind Sorta.
— 01 — the problem
too much leads to too broad
The team were solving for messy homes.
We did tons of researches, spoke to users, and gathered a wide range of insights.
But the more insights we gathered, the more we felt stuck. The team couldn’t answer a simple question:
What exactly are we building—and for whom?
The problem wasn’t lack of data.
It was lack of alignment.
Without clarity, every decision risked being wasted effort.
8 pain point categories, 40+ signals - the information overload was REAL.
— 02 — the move
a three-circle magic
To break the ambiguity, I introduced a simple constraint:
A 3-part molecule model:
Who has the problem
What problem they have
What solution we provide
Each team member created their own version.
Because the space was limited, it forced us to:
Cut vague thinking
Remove assumptions
Get to the core of the idea
We aligned on one molecule.
That became our decision foundation—every product discussion anchored back to it.
Each teammate took 20 minutes to fill out a version of their visioned product.
After voting & discussing, we are aligned on one version.
— 03 — pre-mvp validation
testing without building
Our concept relied on AI guiding people to organize their homes. But that raised a critical question:
Do people actually trust AI to tell them how to organize their space?
If the answer was no, the product fails—no matter how well we design it.
———————————-
Instead of building an MVP, we ran a smoke-and-mirrors test:
Collected user preferences
Manually generated organization plans using AI
Delivered results as if it were a real product
This allowed us to test:
Trust in AI guidance
Perceived usefulness
Impact on efficiency
Result:
Users found the guidance helpful—and trusted it.
That single validation unlocked the next step.
10 users. Real spaces. ChatGPT as the engine. Before a single screen was designed.
Screenshot of partial experiment
An example of one flow
— 04 — THE prioritization
to build or not to build, that is a question.
We had two engineers.
We couldn’t build everything.
So I led feature definition using:
User story mapping → mapping the full cleaning journey
MoSCoW prioritization → defining what must exist vs. what can wait
Every feature had to answer:
Does this reduce cognitive load or save time?
If not, it didn’t make the cut.
Goal: Avoid mental stress & time-wasted caused by decision-making at all stages of the organization process.
User story mapping
MoSCoW matrix
— 05 — THE system
building for speed and scalability
Before designing features, I built a full design system:
Visual language (color, typography, components)
Voice & tone
AI companion character: Sortie
The system allowed us to:
Move faster
Stay consistent
Scale decisions without rethinking everything
Decluttering companion: Sortie
The team created this mascot to build rapport between the product and the users.
Sortie reframes the product from a tool that gives instructions to a companion that reduces stress.
Sortie's original sketches
Voice & Tone
Colors
Typography
Component Library
Screenshot of component library
— 06 — THE product
A feedback-driven system that grows over time
From the beginning, our goal wasn’t just to help users clean.
It was to reduce the mental effort required to make decisions.
But cognitive load doesn’t disappear in a single session.
It compounds over time—and so should the solution.
So we designed Sorta as a learning system, not a static tool.
For onboarding, we built a simple 3-step framework:
How the System Learns
The product continuously adapts through multiple feedback loops:
Before the session
Captures user preferences, goals, and contextDuring the session
Observes behavior: time spent, skipped items, friction pointsAfter the session
Collects feedback, ratings, and outcomes
Each interaction updates the system’s understanding of the user.
Sorta's feedback loop:
And we built features targeting users' pain points along the way, from end to end.
What This Enables
Over time, the product shifts from:
Generic suggestions → Personalized guidance
One-time help → Compounding efficiency
Manual decisions → Assisted decision-making
The goal isn’t automation.
It’s reducing the number of decisions users need to make—without removing their sense of control.
For full product overview, visit Sorta AI's official website: https://www.letsorta.ai/
— 07 — the future
app as the interface. data as the product. home as the platform.
Right now, Sorta helps you decide what to keep.
But every decision you make — keep, toss, donate, reorder — is a data point. And data points, over time, become a model. A model of how you live, what you consume, when things run out, and what your home actually needs.
That's not a decluttering tool. That's infrastructure.
The home as a connected platform
Most smart home products solve for one problem in isolation. None of them talk to each other at the level that matters: what does this household actually own, use, and need?
Sorta sits at that intersection. Because we built the companion relationship first, we have something none of those products have — behavioral context. We know not just what's in your home, but how you relate to it.
That context is what makes every external connection valuable:
When Sorta connects to Amazon, it's not just a reorder button — it's a replenishment engine that knows your consumption rhythm before you do.
When it connects to Facebook Marketplace or a local resale platform, it's not just listing items — it's routing the right object to the right buyer at the right moment, because it knows the item's history.
When it connects to a recycling center, it's not just disposal — it's closing the material loop with data that helps recyclers sort, value, and redirect goods more efficiently.
When it connects to home robotics, the companion becomes the brain. The robot doesn't just clean — it acts on a real model of what the home needs.
The flywheel no one else can build
Every interaction makes the model better.
Every decision you make trains Sorta to need fewer decisions from you next time. That's the loop — and it compounds. After six months, Sorta isn't just helping you declutter. It's quietly managing the lifecycle of everything you own.
The moat isn't the features.
It's the data that accumulates because you trusted the companion enough to keep using it.
No competitor can replicate that data without the relationship that generates it.
Feature prioritization
Narrowed 40+ ideas to a single validated feature set