FinSeek

Inspiration

Inspired by seeing our dads work in risk management at major banks like Wells Fargo and JPMorgan Chase, we’ve heard firsthand about the challenges of detecting fraud while still being unable to stay ahead, forcing an overallocation of resources to cover this risk. FinSeek was born to solve this problem—leveraging AI and predictive machine learning to not just detect fraud, but anticipate it before it happens.

What it does

FinSeek combines three AI models (High-Precision LightGBM, Random Forest, and High-Recall LightGBM) into a 2-of-3 ensemble system that flags fraud only when at least two models agree, achieving over 99% precision with minimal false positives

Beyond detection, FinSeek predicts future fraud trends and calculates prevention budgets using linear regression, exponential smoothing, and moving averages, enabling organizations to proactively manage risk and allocate resources efficiently.

How we built it

We trained FinSeek on real financial datasets using open-source LightGBM implementations, combined with a custom logistic regression model for transaction risk weighting. The backend was developed with Cursor and deployed on Render for proper API integration. Figma was used to design a clean, efficient UI/UX.

Our workflow combined feature engineering and predictive modeling, creating a deployable system that accurately detects and predicts fraud and enables effective allocation of prevention budgets.

Challenges we ran into

We faced challenges at every stage: finding high-quality datasets, building and tuning the models for precision and recall, designing an optimized end-to-end UX, and ensuring our backend deployed via Render could handle requests efficiently.

Accomplishments that we're proud of

We're proud of our LightGBM model integration, as well as our custom logistic regression model and its training. Our UX is also optimized for employee use, developed and integrated through informal interviews with professionals in the field.

What we learned

We learned about machine learning through this project. Using LightGBM for 2/3 models taught us about integrating existing tools into a new project, as we did here.

We also learned the importance of feature engineering and application flow, minimizing total clicks to achieve the tool's purpose as efficiently as possible.

What's next for FinSeek

We hope to fine-tune this application and eventually present it to banking or finance organizations across Texas A&M to gain more insight and advice on how to develop this idea into something truly usable by banks nationwide.

A major aspect of this application that is not yet addressed is overall security. For example, a small but high-impact issue is our current storage system: we use localStorage to store sensitive (transaction information), which is easily scraped by XSS attacks. We plan to implement secure server-side storage, encryption, and authentication protocols to ensure that FinSeek is not only accurate and predictive but also safe for handling sensitive financial data, preparing us for the next step of industry use.