Inspiration
Managing money often means switching between banking apps, credit card statements, subscription lists, and budgeting tools. Even with all that information, it can be difficult to understand what to do next. We created FinPilot to bring a person's financial picture into one workspace and turn financial data into understandable, evidence-backed insights.
What it does
FinPilot is a personal finance prototype where users can explore accounts, transactions, cash flow, and upcoming bills through simulated institution connections.
Its backend supports subscription analysis, debt assessment, goal planning, and financial stress scenarios. A NVIDIA Nemotron chatbot integration provides the foundation for answering financial questions using supporting evidence.
Our demonstration uses fictional financial profiles and simulated connections. The chatbot backend is implemented, with its frontend integration planned next.
How we built it
We built the frontend with Next.js, React, and TypeScript, using reusable components and layouts informed by Figma designs. The backend uses Node.js and SQLite to manage financial records, connections, sessions, and analysis results.
We integrated NVIDIA Nemotron for specialist reasoning and conversational explanations. Financial calculations and access controls remain in deterministic server code. Docker packages the application, while Mailpit supports local email verification.
Challenges we ran into
Keeping account balances, transaction histories, and connection states consistent required careful validation. Integrating the frontend with authenticated APIs also meant handling expired sessions, unavailable services, and browser-origin restrictions.
Another challenge was updating older demo databases without losing existing progress. We developed and tested a targeted repair that preserved financial records and saved connections.
Accomplishments that we're proud of
We built a working financial dashboard backed by persistent data and a password-plus-email verification flow.
We verified login, fresh email codes, session creation, CSRF protection, and logout. We also repaired an existing database while preserving more than 12,000 transactions.
Our analysis backend connects findings to supporting evidence, creating a foundation for financial explanations that users can inspect.
What we learned
We learned that useful financial AI starts with consistent data, clear authorization, and reliable calculations. Explanations become more valuable when users can trace them back to evidence.
We also learned that integration details matter: database migrations, browser origins, API contracts, and persistent storage all affect the complete user experience.
What's next for FinPilot - PNC and Nvidia challenges
Next, we plan to connect the chatbot interface and complete the Agent workspace so users can ask questions, request financial checkups, and inspect supporting evidence.
We also want to expand the planning and scenario interfaces, evaluate Nemotron responses across more financial profiles, improve accessibility, and prepare the application for hosted deployment. Longer term, we hope to explore consent-based financial data integrations.
