Guide-A-Bull

The Intelligent Campus Navigator for University of South Florida Students.

Comment

🤘 Guide-a-Bull AI

The Intelligent Campus Navigator for University of South Florida Students.

Inspiration

As students at USF, we realized that finding a simple form—like the "SSN Letter Request" for a new job or the "Immunization Upload" portal—often feels like a treasure hunt. The university website is massive, link-heavy, and confusing, especially for freshmen who don't know the difference between "OASIS", "Archivum", and "Canvas".

We wanted to build something that cuts through the noise. No searching through five different departments. Just ask a question, and get a direct, step-by-step roadmap to the exact button you need to click.

What it does

Guide-a-Bull AI is a smart web application that acts as a personalized campus concierge.

  1. AI Navigator: Users ask questions in plain English (e.g., "How do I book a study room?"). The app uses Google's Gemini AI to generate a strict, click-by-click script (e.g., "Step 1: Go to LibCal. Step 2: Click 'Reserve Space'.").
  2. Smart Link Verification: It doesn't just guess links. We built a "Verified Dictionary" system that intercepts requests for critical services (Housing, Dining, Parking) and routes users to the live, official portals, guaranteeing the links actually work.
  3. Freshman Shortcuts: It offers one-click access to the 15 most essential portals (MyUSF, Canvas, Bull Runner) so students don't have to bookmark twenty different pages.
  4. Domain Safety: It intelligently filters out results from the "University of San Francisco" (USFCA), ensuring all advice is strictly for USF Bulls (Green & Gold).

How we built it

We built the frontend using React 19 and TypeScript for a robust, type-safe architecture.

  • Design: We used Tailwind CSS to create a custom "Dark Mode" aesthetic featuring the official USF Green (#006747) and Gold (#CFC493), with "glassmorphism" effects to make it look modern and sleek.
  • AI Engine: We integrated the Google Gemini 2.5 Flash API. We spent significant time on "Prompt Engineering" to force the AI to be concise (max 1 sentence per step) and act as an authoritative guide.
  • Routing: We implemented a custom logic layer in geminiService.ts that acts as a traffic controller, deciding when to let the AI search and when to inject hardcoded, verified URLs for reliability.

Challenges we ran into

  • The "Two USFs" Problem: Initially, the AI kept giving us links for the University of San Francisco (USFCA). We had to implement a strict "Negative Constraint" in the system prompt to ban usfca.edu domains and enforce the South Florida context.
  • Link Rot: University sub-pages change URLs frequently. A link to a specific PDF might work today and break tomorrow. We solved this by training the AI to guide users to "Landing Pages" (stable) rather than "Deep Links" (unstable).
  • UI Balance: Making the site look "cool" (Dark Mode/Glass) while ensuring text remained readable and accessible was a constant balancing act. We iterated on font sizes and contrast ratios multiple times.

Accomplishments that we're proud of

  • The "Click-by-Click" Engine: We successfully tuned the AI to stop writing essays and start writing scripts. It no longer says "You can find this on the website..." and instead says "Click the Green Button labeled 'Login'."
  • Official Branding: The app genuinely feels like a USF product. The "Horns Up" logo, the specific hex codes, and the typography match the campus spirit perfectly.
  • Zero-Hallucination Core: We are proud of the "Truth Dictionary" implementation that guarantees 100% accuracy for high-stakes links like tuition payment and housing applications.

What we learned

  • Prompt Engineering is Coding: Writing the instructions for the AI was just as complex as writing the JavaScript. We learned how to use "System Instructions" to effectively "program" the AI's behavior.
  • User Experience Matters: Freshmen don't want technical jargon; they want to know "Where do I click?". Simplifying the language was key.
  • The Power of Verification: AI is great, but hybridizing it with hardcoded logic (verified links) creates a much more reliable product than AI alone.

What's next for Guide-a-Bull AI

  • Voice Navigation: Integrating Gemini Live to allow students to talk to the app while walking to class (e.g., "Where is the financial aid office?").
  • Campus Map Integration: Adding a visual map overlay to show exactly where buildings are located alongside the instructions.
  • Syllabus Scanner: Allowing students to upload a PDF syllabus and ask "When is my first exam?" using Gemini's multimodal capabilities.
Share this project:

Updates