PantherAI – Your FIU Campus Copilot

Inspiration

As FIU students, we noticed that campus resources—syllabi, announcements, financial planning tools—are scattered and often overwhelming to navigate. Students struggle with keeping track of assignments, remembering important deadlines, and managing personal finances, all while balancing coursework and extracurricular activities. We wanted to create a single “copilot” that centralizes these needs and empowers students to succeed both academically and personally.

What it does

PantherAI is a campus copilot for FIU students that brings together three powerful capabilities in one place:

AI-Powered Campus Q&A – Ask questions about campus resources, uploaded syllabi, or notes and get accurate, context-aware answers.
Task Extraction – Upload your syllabus (PDF or even a photo), and PantherAI automatically parses assignments, exams, and deadlines into structured tasks.
Finance Tracking – Upload CSVs of transactions and receive instant summaries of spending and trends, helping students budget smarter.
Modern UI – A clean, responsive, Go-Blue themed chat interface for seamless interaction.

How we built it

Frontend: React + TypeScript with Vite for fast builds and Tailwind CSS for a modern, FIU-inspired UI.
Backend: Flask (Python) with REST endpoints to handle chat, file uploads, syllabus parsing, and finance logic.
AI/ML: Integrated Google Gemini 1.5 Pro for natural language chat and OCR-based document/image ingestion.
Data Processing: PyPDF2 for text extraction, regex patterns for syllabus parsing, pandas + numpy for CSV summaries.
RAG System: Lightweight in-memory retrieval system for context injection into Gemini prompts.
Deployment-ready: Modular project structure with React frontend and Flask backend running locally at localhost:3000 and localhost:8000.

Challenges we ran into

Syllabus Variability: Every syllabus had a different format, so creating regex patterns that captured all assignments and exams was tricky.
ChromaDB Issues: We initially tried vector databases for semantic search but faced dependency issues, so we pivoted to a simplified in-memory keyword search.

Accomplishments that we're proud of

Built a demo-ready full-stack app in the span of the hackathon.
Successfully integrated Gemini AI with retrieval over custom documents.
Designed a clean, responsive UI with a sliding panel and Markdown rendering.
Created a working syllabus parser that can extract structured tasks from multiple syllabus formats.

What we learned

How to integrate large language models into full-stack apps with custom data retrieval.
The importance of UX/UI design in making complex AI interactions approachable.
Regex is both powerful and painful—good patterns can save hours, but edge cases are endless.
The balance between scalability vs. demo simplicity: in-memory solutions worked fast for the hackathon, but we also planned how to expand with databases and persistent storage.

What's next for PantherAI

Persistent Storage: Move from in-memory to a proper database (PostgreSQL/MongoDB).
Authentication: Add user logins with JWT for personalized dashboards.
Enhanced RAG: Upgrade to semantic embeddings with ChromaDB or Pinecone for smarter document search.
Full Finance Features: Expense categorization, charts, savings goals, and financial planning tools.
Mobile App: Expand PantherAI to iOS/Android for on-the-go student use.
Real-Time Updates: Enable WebSocket-powered live task updates and notifications.