FinCopilot

Inspiration

Most students have financial goals but no clear way to know if they're realistic. Budgeting apps tell you where your money went. Nothing tells you what to do next. We wanted to build something that actually helps students make decisions, not just track the past.

What it does

FinCopilot is a conversational finance copilot for students. You chat with it like a financial advisor. As you answer questions about your income, expenses, debt, and goals, it builds a live financial profile in real time. From there it does three things:

Detects issues using a deterministic rules engine (things like spending more than you earn, keeping savings in a non-interest account, or having an unrealistic goal timeline). Each flagged issue is explained in plain language with a severity level.
Simulates scenarios so you can ask "Can I afford to move out in 6 months?" and see a side-by-side comparison of your current vs. hypothetical trajectory across cash flow, debt, and savings. An AI-generated verdict explains what the numbers mean.
Launches a browser agent that researches real options (HISAs, balance transfer rates, housing costs) and brings the results back into the conversation for you to act on.

How we built it

The frontend is built with Next.js, React, and Tailwind CSS with a custom neumorphic design system. The backend is FastAPI with Supabase for auth and persistence. The AI layer uses the OpenAI API for conversation, structured data extraction, and scenario explanations. A Playwright-powered browser agent handles real-world research tasks. We kept a strict separation between AI (language and explanation) and deterministic code (all financial math and rule evaluation).

Challenges we faced

Getting structured financial data reliably extracted from free-form conversation required careful prompt engineering and a second LLM extraction pass after each message. Keeping the live profile sidebar in sync with the chat in real time without race conditions took several iterations. Scoping the browser agent to be useful without being unpredictable under hackathon time constraints was the biggest design challenge of the weekend.

What we learned

How to combine deterministic rule engines with LLM explanation layers so the output is both trustworthy and readable. How to design a product demo flow before writing a single line of code. How to cut scope fast when the clock is running.