Arrowster
Designed and shipped the foundation of Arrowster — a college consulting platform backed by the founding team and board at OpenAI, Duolingo, and Meta — replacing spreadsheet-based workflows with a unified internal and consumer-facing tool, now supporting 1,000+ students.
Led mixed-method research with students and counselors; translated findings into a new product structure that consolidated 20+ features into 6 specialized AI counseling agents
Designed and launched an internal platform for counselors to manage student workflows, consolidating 15+ spreadsheets and 7+ tools into 2 unified tools with AI automation that speeds up the counseling process, reducing manual work by 5+ hours weekly
Integrated AI-powered tools (Cursor, Claude Code) to rapidly prototype and validate concepts
Date
March 2025 - September 2025. In development
The team
1 Product designer, a team of 5 developers
What I did
User research, wireframe, hi-fi prototype, user testing, final product
01
Background: Redefine traditional studying abroad model
Arrowster started with a simple insight
Traditional study abroad agencies rely on salespeople with little real-world experience to guide students through one of the biggest decisions of their lives.
The founders wanted to flip that model. At Arrowster, students are supported by counselors who've actually lived the experience: real students who've studied and worked abroad themselves. People who've been in the same boat and know what it actually takes.
My task
I joined Arrowster at a pivotal moment: they knew customer success was their next priority, but weren't sure where to begin. But they didn't yet have a design process, a clear product direction, or any existing framework to build from.
I was brought in as the sole product designer to map out where the biggest friction points were for students and counselors, and design a unified platform that they could actually rely on day-to-day, starting from zero with no existing design process.
02
First problem: Improving counseling process for in-house counselors
I started by interviewing both students and counselors to understand how they are working to communicate and track students. Through conversations with 20+ students and 3 counselors over two weeks (using in-depth interviews, focus groups, and workflow observations), a clear pattern emerged.
A tangled web of systems
The team was using 15+ spreadsheets with hundreds of tabs to manage leads, student details, and application progress, along with several tools for forms, files, and emails.
Fragmented process
Counselors each had their own system: one used Google Sheets, another preferred Notion. This approach was consuming 7+ hours weekly per counselor on data entry and status update alone.
"Everyday when I wake up I just have this constant fear that I might have missed some students' important deadlines" - Hayden, counselor
The previous process
The team was juggling 15+ Google spreadsheets with hundreds of tabs to track leads, student information, and application statuses. On top of that, they used Tally for form submissions, Google Drive for storing applications, and Zapier for basic email automation.
Image: Previous counseling process of Arrowster, with 15+ spreadsheets and 7+ other supporting apps
My proposed process
As Arrowster looked to bring on more students and counselors, these inefficiencies would only compound. The core solution is to build a unified system to automate and consolidate counselor workflows into a simplified product flow.
Image: Consolidated everything into a simplified flow, using Zoho for CRM and Product for onboarding students and application support
Solution
I decided to build a system for counselors and students to interact with each other, and involve Sophie AI deeper into their studying abroad journey. Because each journey is unique to the student's experience, background, and goals, I tried to simplify the working process as much as possible, rather than a carefully planned step-by-step process like before
02
Second problem: College counseling tool for students
The current process
Sophie AI was designed to guide students through the application process. However, its checklist format often caused more confusion than clarity: students were directed to complete a checklist of over 20 steps, covering everything from career guidance to visa applications.
Before: The list covered everything from career guidance and major selection to test prep, essays, and visa applications—often 20+ tasks at once.
Students reported feeling overwhelmed and unsure which steps were relevant to their specific situation. Most didn't use every feature and didn't know where to start.
Some iterations
As we were in the MVP phase, I scoped the problems into two main areas:
Enhancing the product experience: Improving navigation to crucial touch points, addressing critical UX issues while ensuring the design is scalable for future features,
Simplifying user interaction: Reducing cognitive load and friction for students when interacting with Sophie AI, allowing more personalization for students
I sketched out some ideas for the information architecture and discussed them with the team to check their expectations and spot any potential red flags.
From a technical standpoint, one requirement is to use multiple agents (or specialized models) instead of a single combined one, since a large token load can increase latency and slow down response times. This approach allows each agent to focus on a specific task, making the system more scalable, reliable, and easier to maintain.
Solution
Based on in-depth interviews and the word cloud of students’ main concerns about studying abroad from online discussions, I merged the AI agents into six key angles most relevant to their needs. After completing the onboarding questions, we will direct the student to their current situation. The 7 angles are:
Choosing major and Career guidance
School selection and scholarship recommendation
Planning extracurricular activities
Preparing for tests
Writing essays
Visa application
I also redesigned the information architecture and streamlined navigation, making it easier for students to find AI agents and enhancing profile personalization so they have a more tailored experience.
04
Results
Fixing both sides of the problem is what made the impact real. Cleaning up the counselor workflow without fixing the student product, or vice versa, would have only solved half the equation.
For counselors: Consolidating 15+ spreadsheets and 7+ tools into 2 unified platforms saved 5+ hours of manual work per counselor each week.
For students: Replacing the overwhelming 20-step checklist with 6 focused AI agents gave students a clearer, more personal starting point. Instead of feeling lost in a generic list, they could engage with the parts of the journey that actually mattered to them right now.
For the business: The platform now supports 1,000+ students and internal teams, and contributed to Arrowster securing $5M+ in investment, a signal that the infrastructure built during this phase gave the company the foundation it needed to grow with confidence.
05
My lesson
One of the biggest takeaways from this project is how much I grew in managing work inside a small team and building a design process from scratch, when there was nothing to inherit. Finding the right tools and rhythms for communication is never a one-size-fits-all thing. It takes observation, patience, and a lot of trial and error.
One of the focus group interviews we have with high-school students
But more than the tools themselves, I learned that design processes don't belong solely to designers. The most valuable thing a designer can do in a small team is create a space for meaningful conversations about vision, goals, and real challenges, not just hand over screens.
