Opendoor

Reducing barriers to essential services with the help of AI

Summary

Opendoor is a mobile app designed to help people find and access local shelter, meals, and support services in minutes, cutting through the fragmented, outdated resources that used to make this search take hours.

My Contribution

I co-founded Opendoor, leading the entire end-to-end design as we reimagine the concept as a nationwide, AI-powered resource aggregator. I've led designs for 3 core user flows across 10 screens for iOS and Android.

Timeline

From May 2025 to January 2026

Team

1 Product Manager

3 Designers

2 Engineers

Skills

User Research

Interaction Design

Prototyping

Usability Testing

Tools

Figma

UserTesting

Claude Code

At a Glance

From a side project to award-winning, reimagining how people access essential resources nationwide.

Key Highlights

$10k won

as the Grand Prize Winners of the Kevin Xu Innovation Challenge

50 Users

tested the prototypes, 92% said they'd use OpenDoor if deployed

Selected

Builder cohort, 2025 Harvard Applied AI Grid Incubator and Amazon Web Services

5+ shelters

expanding early pilots across the Bay Area and Boston

Overview

From a side project to award-winning, reimagining how people access essential resources nationwide.

Key Highlights

$10k won

as the Grand Prize Winners of the Kevin Xu Innovation Challenge

50 Users

tested the prototypes, 92% said they'd use OpenDoor if deployed

Selected for

2025 Harvard Applied AI Grid Incubator and Amazon Web Services

5+ shelters

expanding early pilots across the Bay Area and Boston

Search and filter

Categorized listings for housing, legal aid, recovery, and job search, with filters for open-now status, walk-in availability, and accessibility needs, helping people in crisis narrow results in seconds, not another site to dig through.

Everything a user needs to know

Each listing shows eligibility requirements, what to bring, and services offered, along with hours, availability, and directions, giving users everything they might need to confirm they qualify for that specific resource and get there, without wasting a trip.

A second way to search using AI

Alongside the standard browse-and-filter experience, users can describe their situation in plain language (what's going on, what they need) and the AI interprets that to surface the most relevant resource, no prior information required.

The Problem

Gaps in access prevent people who need them most from connecting with available local resources.

This problem surfaced while auditing the Good Samaritan Shelter website: even after finding the right service page, users had to scroll through 4 screens before reaching eligibility requirements, one of the first things people need to know before calling or visiting. That audit became the foundation for Opendoor, rebuilt as a mobile-first experience designed around clear information hierarchy from the ground up.

User Research

Our core user on average had to spend 20+ min searching 4+ sites to find a single relevant service.

Through 13 stakeholder interviews and surveys with 50+ program respondents ( conducted during our initial research for Good Samaritan), two patterns surfaced consistently. Users were spending time searching across 4+ fragmented, outdated sites just to find one relevant service, and when they arrived, critical information like eligibility requirements, walk-in hours, and what to bring was either buried or missing entirely.

Key Insights

48%

Reported a wasted trip due to missing or unclear eligibility information

20+ minutes

Minutes spent searching, on average, per service found

4+ sites

Sites were visited on average before finding one relevant service

87%

Rely solely on a Lifeline-subsidized phone for internet access

Search and filter, designed around real constraints

Categorized listings for housing, legal aid, recovery, and job search, with filters for open-now status, walk-in availability, and accessibility needs, helping people in crisis narrow results in seconds, not another site to dig through.

Everything a user needs to know, in one view

Each listing shows eligibility requirements, what to bring, and services offered, along with hours, availability, and directions, giving users everything they might need to confirm they qualify for that specific resource and get there, without wasting a trip.

Ask OpenDoor: a second way to search using AI

Alongside the standard browse-and-filter experience, users can describe their situation in plain language (what's going on, what they need) and the AI interprets that to surface the most relevant resource, no prior information required.

Competitive Analysis

Current solutions fail to serve people navigating financial strain with local, contextual needs.

The patterns from our Good Samaritan research pointed to a gap larger than one shelter's website. Existing directories like 211 and FindHelp aggregate resources but are hard to navigate, while general tools like Google Maps and Yelp are easy to use but aren't built for the specific, often urgent needs of vulnerable populations. Opendoor was built to sit in the gap between the two: a resource finder that's both comprehensive and genuinely easy to use under pressure.

Product Differentiation

Our Bet: Conversational AI to reduce staff burden and personalize support as needed.

To maximize impact with limited resources, we categorized potential features into an impact matrix, prioritizing AI search and offline access as critical 'Big Bets' while cutting low-value friction like forced account creation. This helped us focus our time on the features most likely to improve access for users, rather than spreading resources across every possible idea.

Ideation

Using AI to speed up early ideation

To explore direction quickly, we used AI to help generate and iterate on multiple early concepts in parallel, ranging from a simple hotline-first layout to a guided chat assistant, a map-based explorer, a categorized directory, and an action-plan checklist. This let us compare structurally different approaches before committing to a direction.

Information Architecture

Defining the core MVP user flow by identifying what to prioritize and intentionally cut.

Before designing any screens, we mapped every possible user action and narrowed the core experience down to three core MVP flows: the homepage, service details, and conversational AI search, with supporting features like offline access, bookmarks, and accessibility settings built around them.

Key Design Decisions

Evaluating design tradeoffs to find the most intuitive path for users in high-stress situations.

For each key screen, we designed multiple variations rather than committing to a single layout early, weighing the trade-offs of each against real constraints, cognitive load, scannability, and how caseworkers and users actually needed to act on the information. This side-by-side comparison helped narrow toward directions that reduced friction instead of just looking clean.

Filter-First Layout

Has a map/list view toggle (visual search vs. list search)

High cognitive load (e.g wall of controls/filters)

Organized by Categories

Establishes a clear, predictable navigation path

No primary actions like "Call" or "Directions"

Guided-Discovery Layout

Users can discover services through categories or browsing

Users can easily scan for name, status and primary actions

Expanded Card Pop-Up

Groups key information into logical, collapsible sections

Caseworkers noted that important details are buried

Single-Column Details Page

Has a full page view for expanded service details

Still is hard to scan for our users

Easy-to-Scan Details Page

Builds users' trust with an entrance photo, confirming the location is safe to approach.

Uses progressive disclosure by hiding secondary info

Pre-Filtered Smart Search

Relevant filters like sort by "Distance" and "Type of Help"

Users think "Type of Help" UI is complex and unclear

Requires selecting filters before seeing any results

Integrated Search

No complicated pre-search filter barrier

Accidental conflicting touch targets for many users

The service card competes with the search dropdown

Conversational AI Search

Separates search into its own view

Uses progressive disclosure by hiding secondary info

Positions AI as a conversational path to an answer

Design Systems

Building out a WCAG 2.0 Accessible Design System

Opendoor is a mobile platform that helps 5,000+ vulnerable residents in Santa Barbara County find and access local shelter, meals, and support services in minutes, cutting through the fragmented, outdated resources that used to make this search take hours.

Final Prototype

Over the course of 8 weeks, we designed 10+ screens, 26+ components, built for Android and iOS.

Reflection

From problem ambiguity to decisive product bets and shipped experiences.

Early on, I focused heavily on information architecture. But talking with caseworkers made me realize users needed upfront visual proof, like entrance photos and live bed availability, to feel safe approaching a shelter. Trust is an accessibility requirement.

Without strict requirements or a single "standard" user persona, waiting for perfect data would have stalled us. Learning to make opinionated, hypothesis-driven bets allowed us to put real prototypes in front of residents faster.