CampusCoin

Sign Up Screen
Dashboard
Manage Finances
Campus Navigator
Settings

Inspiration

As computer science students juggling coursework, coding projects, and hourly campus jobs, we realized something fundamentally broken about personal finance apps: they are built for 30-year-olds with stable salaries.

College students live in a volatile financial reality. We have massive tuition bills, lump-sum financial aid, and variable-hour jobs. The eureka moment hit us mid-hackathon: academic stress directly causes financial stress. When midterms hit, students work fewer hours, and their income plummets. We realized that tossing a standard AI chatbot at this problem wasn't enough. Students don't want to chat about being broke; they want an engine that predicts the shortfall before it happens and tells them exactly how to fix it.

What it does

CampusCoin is a predictive financial engine built for the student lifecycle.

Instead of just tracking past expenses, it projects a 180-Day Financial Runway so students know exactly when they might run out of funds. Our killer feature is Academic-to-Financial Inference. A user simply drags and drops their course syllabus or Canvas schedule into the app. Behind the scenes, our AI extracts major academic events (like finals week), mathematically infers the expected drop in the student's campus job hours, and instantly recalculates their 180-day runway.

Instead of a chat window, the AI generates "Actionable Insights." These are native UI cards that give 1-click solutions, like: "Your Comp Sci midterms next month will drop your income by $120. Click here to stash $15/week starting today to cover rent."

How we built it

We built a production-lite architecture optimized for speed and inference:

Frontend (The Face): A blazing-fast Single Page Application (SPA) built with React and Vite. We used Tailwind CSS for a polished, accessible UI and Recharts to render the interactive graphs.
Backend & Infrastructure (The Engine): We deployed a serverless Python/FastAPI backend on Modal. To bypass the headache of database migrations during this 36-hour sprint, we utilized Modal Dicts as a highly scalable Key-Value store to manage user sessions and state syncs.
The "Invisible AI" (Gemini + Supermemory): We used the Gemini 3.0 Flash API for multimodal ingestion to read PDFs/images of syllabi and output strict, structured JSON. We then integrated the Supermemory API to persistently store these academic bottlenecks and user financial goals across sessions.
Financial Data: We integrated the Capital One Nessie API to simulate real-time bank syncing, deposits, and recurring bills.

Challenges we ran into

Our biggest hurdle was the classic hackathon pivot. Halfway through the event, we realized our original plan, a financial advisor chatbot, was extremely repetitive. We completely scrapped the chat interface and re-engineered our Gemini backend to operate as an "invisible" agent. We forced the LLM to consistently return complex, perfectly formatted JSON to populate React UI widgets (instead of just generating conversational markdown). This required intense prompt engineering and strict Pydantic validation.

Additionally, calculating the dynamic 180-day runway required some careful state management. We had to ensure the daily projected balance $B_t$ accurately reflected the base cash $B_0$, plus the sum of variable active income streams $I$, minus recurring expenses $E$, adjusted by the AI's inferred hour reductions $A$:

$$B_t = B_0 + \sum_{i=1}^{t} (I_i - E_i - A_i)$$

Accomplishments that we're proud of

The Pivot: Successfully transforming the app from a reactive "wrapper" into a proactive, predictive inference engine in under 12 hours.
True AI Integration: We are incredibly proud of hiding the AI from the user. There's no chat box; the AI simply does the heavy lifting in the background and makes the UI feel like magic.
UI/UX: We built an interface that actually lowers anxiety. By translating complex financial data into simple status badges ("Caution", "On Track") and 1-click actions, we made financial wellness accessible.

What we learned

UX > AI Gimmicks: Users don't want to converse with an AI to solve their problems; they want the AI to autonomously structure the solution for them.
Infrastructure Matters: Modal's serverless ecosystem allowed us to deploy our Python backend and handle heavy Gemini inference tasks without worrying about local environment setups or server crashes.
Startup Mindset: We learned how to frame a hackathon project not just as a cool script, but as a viable B2B2C business model.

What's next for CampusCoin

We are treating CampusCoin as a startup-ready product. Our next technical steps include migrating our Modal Dicts to a robust relational database (like PostgreSQL) and swapping Nessie for Plaid to support live bank accounts.

Business-wise, we plan to pursue a B2B Enterprise Model: selling licenses directly to universities. Universities can offer CampusCoin Premium to their student body as a retention tool, helping catch students before financial crises force them to drop out.