Pathway

Inspiration

One of us is an international student and the other is first-gen. Coming into university, neither of us had parents who knew what a fellowship was, what research experience meant for our futures, or how to navigate the maze of opportunities that college quietly offers to those already in the know.

We wasted tons of ambition feeling lost. We needed a roadmap.

Pathway is built to solve this. Not just for us, but for every student who enters college without a built-in support network - no older siblings who've been through it, no family or friends in the industry, and no mentors pointing them toward the right doors. Pathway can be all of that and more. We wanted to build the friend, planner, and advisor we wish we'd had.

What It Does

Pathway is an AI-powered college roadmap advisor. Students share a bit about themselves including their year, interests, goals, and confidence in their current path. Claude uses these responses to generate a personalized, visual roadmap of opportunities like fellowships, research positions, campus involvements, and more. Each node connects students directly to the resources they need and helps them take actions like drafting outreach emails to real UCLA contacts.

It's a living, path-dependent journey that evolves with every choice the student makes. It helps them organize and achieve their tasks in one place.

How We Built It

• Frontend: Next.js 16 (App Router) + React 19 + TypeScript, styled with Tailwind CSS v4 and shadcn/ui
• Visualization: Custom SVG tree layout engine with RoughJS for a hand-drawn, approachable aesthetic — intentionally designed to feel like a whiteboard session with a mentor, not a corporate dashboard
• AI Core: Anthropic's Claude (claude-sonnet-4-6) via the Claude API, with prompt caching on the system prompt for cost efficiency and consistent behavior
• Data: A curated corpus of real UCLA opportunities with verified contacts, used as grounding context for Claude's suggestions
• State: Zustand for client-side state; sensitive profile data kept strictly in-memory (never persisted to localStorage)

Every task in the student's path is generated by Claude reasoning over the student's full accumulated journey, not just their latest input.

Challenges We Faced

Hallucination and trust. Because our users are vulnerable students making real decisions, a fabricated opportunity can be extremely harmful. So, we built a semantic validation layer that silently drops any opportunity IDs Claude generates that don't exist in our verified corpus.

Prompt injection. Students enter free-form text. We implemented input guards to prevent adversarial inputs from hijacking Claude's behavior which is a intentional ethical design decision.

Balancing personalization with safety. We wanted Claude to feel like a knowledgeable friend, not a liability. That meant being deliberate about what data we store, what we surface (only verified opportunities), and how we frame outputs (as suggestions, not guarantees).

What We Learned

Building for underserved users raises the ethical bar. What we build has to be trustworthy, transparent, and genuinely useful to people who can't afford to be misled. That shaped every technical decision we made.

We also learned that Claude is extremely powerful as a reasoning layer to hold context, adapt to a student's unique path, and generate guidance that feels personal.

What's Next

Pathway is built mainly for UCLA today, but the architecture is designed to generalize to any university, any student population, any corpus of opportunities. The students who need this most are everywhere.

Built With

• anthropic-claude-api
• next.js
• react
• roughjs
• shadcn/ui
• tailwind-css
• typescript
• zod
• zustand