AuraVault

Inspiration

Tracking finances is fundamentally broken. Most people either use complex, boring spreadsheets or apps that require tedious manual entry. I wanted to build something that feels less like a basic calculator and more like having an elite hedge fund manager living in your pocket. I built AuraVault AI, a B2C web application to make financial tracking effortless, intelligent, and highly secure.

What it does

AuraVault AI is a full-stack financial dashboard that completely automates wealth management:

The AI Advisor: A fully context-aware chat interface powered by Gemini 3.5 Flash. It knows your live balance, burn rate, and recent transactions, acting as a real-time behavioral finance expert.
Auto-Vision Extraction: Users can simply upload an image of a receipt or bank statement. The backend passes the image to Gemini Vision to automatically extract the merchant, amount, date, and category, securely saving it to the cloud without typing a single number!
Live Cloud Syncing: All transactions are stored in an AWS DynamoDB ledger, instantly reflecting across the responsive Next.js dashboard.

How I built it

I architected a robust, decoupled full-stack application:

Frontend: Built with Next.js, React, and Tailwind CSS. I implemented a fully mobile-responsive dark mode UI and designed a frictionless, custom "God Mode" bypass so hackathon judges can immediately test the platform without hitting login walls. It is deployed globally on Vercel.
Backend: I built a custom RESTful API using Python and Flask, deployed via Gunicorn on Render.
Database & AI: The backend connects to AWS DynamoDB for high-speed CRUD operations, and utilizes the google-generativeai SDK to process document OCR and conversational AI.

Challenges I ran into

Building a true full-stack app in a hackathon isn't easy! I had to overcome tricky CORS issues between the Vercel frontend and Render backend. The biggest hurdle was a massive 11th-hour architectural pivot: I had to surgically decouple and remove a third-party authentication system across my entire stack to ensure a completely frictionless experience for the judges. I also spent hours perfecting a custom CSS translation matrix to ensure the desktop sidebar transitioned into a flawless off-canvas drawer for mobile users.

Accomplishments that I am proud of

Getting the Gemini Vision model to accurately parse raw receipt images and instantly turn them into structured JSON that updates the AWS database.
Seeing that full pipeline work from a frontend button click to a live database update was incredible!
I am also incredibly proud of surviving and achieving a clean, warning-free Vercel production deployment.

What's next for AuraVault AI

I plan to add live multi-currency conversion APIs, recurring subscription detection, and predictive forecasting to warn users if their current burn rate will empty their vault before the end of the month.