(L)Earning

Inspiration

Our team includes international students and members preparing to study abroad, which made travel spending a shared challenge. Transportation costs often feel small in the moment but add up quickly across rideshares, transit, and fees. We noticed that most finance tools show totals without explaining the patterns behind them.

This inspired us to build a system that does not just track travel expenses, but explains why they happen. Our goal was to turn raw transaction data into clear, educational insight that helps students and travelers understand how convenience and planning shape their financial outcomes.

What it does

(L)Earning is an educational fintech platform that transforms travel and transportation transactions into structured insight rather than simple summaries.

At its core are three integrated layers:

Objective Analysis Engine A rule-based computation layer that converts raw transactions into financial metrics such as category breakdowns, daily spending timelines, peak spending days, fee totals, convenience versus transit ratios, spending rhythm, and transaction fragmentation indicators.

Subjective Educational Analysis (AI Layer) A prompt-driven AI layer powered by Google Gemini that turns structured metrics into short, educational explanations focused on financial behavior, convenience tradeoffs, and cost awareness, rather than prescriptive advice.

Interactive Chatbot Interface A conversational system that combines objective data, prior insights, and conversation context to let users explore “why” and “what-if” scenarios through personalized, context-aware financial education.

The platform is implemented with a FastAPI backend and a modern frontend built using Next.js, React, and TypeScript, with realistic transaction data provided through the Capital One Nessie API.

How we built it

We designed (L)Earning as a layered full-stack system where each component has a clear technical role and clean interface.

On the backend, we built a FastAPI service that integrates with the Capital One Nessie API to simulate realistic financial data flows. Transactions are normalized into a consistent schema and passed into a deterministic objective analysis engine that computes structured metrics such as category breakdowns, daily spending timelines, fee totals, convenience ratios, and spending rhythm indicators. These results are exposed through versioned REST endpoints that act as the single source of truth for both the AI system and the frontend.

For the AI layer, we implemented a prompt-driven pipeline using Google Gemini. Instead of passing raw transactions, we pass structured financial outputs into carefully constrained prompts. This ensures the model behaves as a financial educator, generating explanations grounded in real data rather than generic summaries or prescriptive advice.

On the frontend, we built a React and Next.js interface in TypeScript that consumes the backend’s API endpoints to display objective metrics, educational insights, and a live chatbot. The UI is structured around a clear progression from data → explanation → interaction, allowing users to first see what happened, then understand why, and finally explore their own questions.

Throughout development, we used shared data contracts and strict response schemas to keep the frontend, backend, and AI layers aligned, which helped us move quickly without breaking system boundaries.

Challenges we ran into

We built the system using several frameworks that were new to parts of our team, which required rapid learning and architectural iteration. Separating deterministic analysis from AI interpretation was particularly challenging and forced us to rethink how data should be structured and validated across layers.

We also spent significant time refining prompt design to ensure the AI produced consistent, educational responses rather than generic or prescriptive output. Debugging issues across frontend, backend, and API boundaries highlighted the importance of strict data contracts and clear system interfaces.

Accomplishments that we're proud of

Delivering a complete, working full-stack system in 36 hours, including backend services, AI integration, and a modern frontend

Zero merge conflicts, achieved through clear task ownership, consistent coding standards, and disciplined branch management

Designing and executing a central technical concept, where deterministic financial analysis and generative AI work together instead of independently

Building an AI system that teaches financial concepts rather than simply summarizing or advising

Maintaining clean system boundaries between data ingestion, computation, interpretation, and presentation under tight time constraints

What we learned

We moved beyond building a standard CRUD application and designed a layered, API-driven system for financial data processing. We learned how to structure a REST backend that ingests transaction data, performs deterministic analysis, and exposes clean, machine-readable metrics.

We also gained hands-on experience applying AI as an interpretive layer rather than a replacement for logic. By engineering prompts around structured financial outputs, we learned how to guide a model to produce consistent, educational explanations grounded in real data.

As a team, we practiced designing system boundaries between data ingestion, objective computation, AI interpretation, and frontend presentation, which strengthened our collaboration, debugging workflow, and architectural thinking.

What's next for (L)Earning

Our next step is to deploy the platform for real-world use, allowing students and travelers to interact with the system beyond a demo environment.

We plan to expand bank and financial service integrations to move from simulated data toward secure, real transaction pipelines. This will allow the objective analysis engine to operate on live financial behavior while preserving user privacy and data integrity.

On the security side, we aim to implement robust authentication and authorization layers, including user accounts, role-based access control, and encrypted API communication. This will prepare the platform for broader adoption in educational and organizational settings.

Long-term, we see (L)Earning evolving into a travel-focused financial literacy platform that combines real-time analytics, multilingual AI education, and planning tools that help users understand the financial impact of their transportation choices before, during, and after a trip.