CometPath AI

Inspiration

Every semester at UTD, the course registration period brings a wave of anxiety. Students waste hours cross-referencing PDFs, prerequisite chains, and professor ratings just to figure out what classes to take next. One simple scheduling mistake can delay graduation by an entire year, and academic advisors are often too booked to help everyone. We wanted to build a tool that takes the guesswork out of graduation and lets students focus on their future, not their spreadsheets.

What it does

CometPath AI is an intelligent, automated academic advisor tailored specifically for UTD students. Instead of manually building schedules, students simply upload their unofficial UTD transcript as a PDF. Our app instantly parses their academic history and generates an optimized, multi-semester roadmap.

Prerequisite Checking:

It automatically verifies eligibility for future courses. Workload Balancing: It estimates the difficulty of your selected courses to prevent burnout. Campus Copilot: It acts as a career matchmaker, algorithmically recommending highly relevant UTD student organizations, hackathons, and campus events based on the student's specific career track (e.g., Cybersecurity, Machine Learning).

How we built it

We built the entire application using Next.js 14 and React for a blazing-fast, server-rendered experience, styled with a custom Tailwind CSS design system. For the backend, we processed and synthesized over 600MB of real UTD course and section data into lightweight, queryable JSON files. Instead of a complex cloud database, everything runs natively within Next.js API routes. For the AI capabilities, we integrated the Google Gemini 2.5 Flash API alongside pdf-parse to accurately extract and understand raw transcript PDFs, translating unstructured text into a structured student profile.

Challenges we ran into

Our biggest challenge was accurately parsing messy, unstructured PDF transcripts into reliable data. Every transcript is formatted slightly differently, so we had to write robust multi-part prompt engineering to ensure Gemini could consistently extract completed courses without hallucinating. We also faced rate-limiting issues when trying to use LLMs to dynamically search for campus clubs, which we successfully solved by building our own local, algorithmic keyword-matching engine for the Campus Copilot feature.

Accomplishments that we're proud of

We are incredibly proud of the multi-semester "Roadmap" visualizer. Taking a flat list of JSON courses and turning them into an interactive, chronological graph that visually proves a student can graduate on time was a huge technical and design win. We're also proud that the app requires zero manual data entry—uploading a PDF and instantly seeing a 4-year plan feels like magic.

What we learned

We learned a massive amount about working with Next.js App Router, particularly handling server-side API routes and managing large local datasets efficiently. We also learned practical techniques for prompt engineering and working with the Gemini API to reliably extract structured JSON from noisy document data.

What's next for CometPath AI

In the short term, we want to integrate real-time UTD CourseBook API data so students can see live seat availability for the sections we recommend. Long term, we envision integrating a social feature where students walking similar paths (like two freshmen both doing the ML track) can connect and form study groups directly through the Campus Copilot!

Built With

• google-gemini-api
• next.js
• react
• tailwind-css
• typescript