Husky Assistant

Inspiration

As students, we experienced how overwhelming and confusing it can be to navigate campus resources. Many students—especially first-year and international students—struggle to find the right support when they need it most.

Additionally, campus offices and advisors are not always available, making it difficult to get questions answered in real time. This often leaves students stuck, unsure of where to go next.

We wanted to design a solution that makes accessing help feel simple, immediate, and human-centered—specifically tailored to UW Bothell students.

What it does

Husky Assistant is an AI-powered platform designed specifically for UW Bothell that helps students quickly find the campus resources they need.

Instead of navigating multiple websites or waiting for office hours, students can ask natural questions like:

“I’m stressed—what resources can help me?” “Where can I find advising support?”

The system responds instantly with relevant, easy-to-understand guidance and direct links to UWB resources.

How we built it

Stack: React (Vite) front end for chat and resource cards, Flask back end with an /ask endpoint, and a Groq LLM for replies.

Core idea: The model does not invent links or event times. It uses MCP (Model Context Protocol): a small MCP server (Python) exposes tools (e.g. campus events on Gather, advising, career, wellness). The LLM calls a tool when needed, gets back JSON, and turns that into a short markdown answer with real URLs and details.

Why MCP: It keeps the chat logic separate from each data source, so we can add or change integrations without rebuilding the whole app, and answers stay tied to actual tool data instead of guesswork.

Challenges we ran into

Designing responses that feel helpful and human, not robotic
Structuring campus-specific information clearly by using MCP's using the API keys
Ensuring accessibility and ease of use for all students
Debugging backend and frontend integration

Accomplishments that we're proud of

Built a solution tailored specifically to UWB students
Addressed a real gap in campus support accessibility
Created a clean, user-friendly interface
Developed a working full-stack prototype within hackathon time

What we learned

How to design solutions around real user pain points
Importance of accessibility and immediacy in student support
Building systems that enhance human experience, not replace it
Full-stack development and integration

What's next for Husky Assistant

MCP as a service :

Offer a hosted MCP layer (or small SDK) so student orgs, departments, or event hosts can register their site or event feed, map a few fields (title, time, link, contact), and have those surfaces searchable by the assistant—without every team rebuilding integrations.

Broader UW integration :

Expand official UW / UWB sources (HuskyLink, key .edu pages, calendars where allowed) through more MCP tools and clearer data-use and attribution, so answers stay accurate, link-backed, and maintainable as the University’s web presence changes.

Safer, scalable rollout :

Add auth and tenancy (who can register what), rate limits, and a review path for third-party onboarded content so the assistant stays trustworthy as it grows beyond a hackathon demo.

Institutional LLM for production :

Work with the college / central IT to use officially approved LLM options for the backend (e.g. campus agreements, API access, or credits), so the assistant can run on sustainable, policy-aligned language models instead of ad hoc personal keys—making it realistic to deploy and maintain for real students.