Project LUMEN
AI layer to manage finance
The problem Project LUMEN solves
Our platform automatically reads financial emails—bank alerts, UPI receipts, invoices, bills—and converts them into clear, real-time insights. Users get automatic expense tracking, category-wise spending, pie charts, and monthly trends without manually entering anything.
It also analyses money behaviour to detect unusual spending, hidden subscriptions, or potential fraud. Users receive proactive reminders for bills and recurring purchases.
A unique feature is the wishlist affordability predictor, which tells users exactly when they’ll be able to buy something they want based on their cashflow and saving habits.
Overall, it makes personal finance effortless, organized, and safer—running silently in the background and helping users make better financial decisions.
Challenges we ran into
A major hurdle was integrating the LLM with our database in real time. Running the model locally on the same system caused performance drops, slow responses, and unstable data syncing.
To solve this, we hosted a lightweight LLM on a separate laptop over the local network and connected our main website to it. This split the load, prevented resource conflicts, and made LLM ↔ database communication much more stable.
By separating the inference node and the web app, we achieved smooth processing, faster responses, and reliable structured data storage.
Tracks Applied (1)