Inspiration
Financial stability isn't always what it looks like on paper.
Someone can consistently earn more than they spend and still be just one unexpected expense or income disruption away from financial difficulty. We wanted to explore how banks could identify these vulnerabilities before they become serious problems.
While analyzing 4,500 anonymized bank accounts and over one million transactions, we discovered that 21.6% of accounts with a positive typical monthly surplus would have less than six months of financial runway if their inflow dropped to 55%.
That finding inspired Tangerine Mission Control.
We wanted to go beyond traditional financial dashboards that simply report balances and spending. Our goal was to create a platform that helps banks understand a customer's financial resilience, simulate unexpected life events, and develop practical strategies grounded in their actual financial circumstances.
What it does
Tangerine Mission Control is an AI-powered financial resilience and decision-support platform designed for bank employees.
The platform follows three core steps.
- Understand the customer
By entering an account ID, a bank employee can access a comprehensive financial profile, including typical income, spending, fixed obligations, cash reserves, historical cash flow patterns, and potential financial vulnerabilities.
- Simulate unexpected financial events
Employees can describe a financial situation in everyday language, such as losing part of their income, changing jobs, facing an emergency expense, or experiencing multiple financial shocks simultaneously.
Gemini interprets the description and converts it into structured financial parameters, including amounts, timing, and duration. Employees can also manually adjust these inputs for greater precision.
- Generate a personalized financial response
The platform uses Jev or Gemini to help select an appropriate response from eight financial planning modules covering income disruptions, cost pressures, emergency liquidity, debt obligations, recovery, and surplus allocation.
A deterministic financial engine then calculates projected cash flow, financial runway, the timing of potential cash shortages, additional funding requirements, and the impact on emergency savings and investment planning.
The platform also allows users to compare financial strategies while identifying recommended priorities, outstanding information, and next steps.
Rather than automatically recommending unrealistic spending cuts or investments, Tangerine Mission Control distinguishes between financially supported actions and options that require further verification.
How we built it
We began by analyzing the anonymized financial dataset provided for the Tangerine challenge, containing 4,500 accounts, approximately 1.06 million transactions, and 682 loan records.
Using Python and Pandas, we cleaned, explored, and transformed the financial records into interpretable customer-level indicators, including income stability, cash flow variability, reserve coverage, and historical financial stress.
We then developed a financial planning framework consisting of eight reusable response modules, allowing different combinations of financial shocks to be evaluated without relying on a fixed list of predefined scenarios.
For AI integration, we incorporated Gemini for natural-language scenario interpretation, with Jev and Gemini available for financial response classification.
We intentionally separated AI-assisted interpretation and strategy selection from financial arithmetic. The financial calculations are performed by a deterministic Python engine with validation rules designed to prevent unsupported assumptions and misleading recommendations.
For the application, we used FastAPI, JavaScript, HTML, and CSS to create an interactive platform for bank employees. The interface brings together customer profiles, financial visualizations, scenario analysis, personalized assessments, and review management.
We also developed independent financial validation cases to compare the engine's calculations against expected results, including combined shocks, delayed payments, debt deferrals, and incomplete financial information.
Challenges we ran into
One of our biggest challenges was ensuring that our recommendations were not only mathematically correct but also financially responsible.
The dataset contains extensive transaction histories, but it doesn't tell us everything about a customer's current circumstances. For example, historical withdrawals don't necessarily indicate discretionary spending, and a historical account balance isn't proof of currently accessible cash.
We had to design our system around these limitations rather than assume information that wasn't available.
Another challenge was handling complex financial shocks involving multiple changes at different times. A customer might experience reduced income, an immediate expense, and a delayed payment within the same period. A simple monthly average could overlook a critical cash shortage.
We addressed this by developing a financial projection engine that tracks cash movements over time and identifies the first negative cash checkpoint.
We also had to ensure that AI-generated classifications couldn't override financial calculations or introduce unsupported actions.
Finally, connecting data analysis, AI models, financial logic, and an intuitive interface into one coherent product required careful coordination across our team.
Accomplishments that we're proud of
Our biggest accomplishment is building a system that goes beyond identifying financial vulnerability to evaluating what could realistically be done about it.
One of our most meaningful findings was discovering that 714 of 3,309 accounts with a positive typical monthly surplus would have less than six months of runway under a hypothetical scenario where inflow drops to 55%.
We're also proud of developing a flexible scenario engine capable of representing combinations of financial changes, including recurring income reductions, unexpected expenses, and payment timing disruptions.
Our eight-module financial planning framework allows different financial circumstances to lead to different priorities rather than generic recommendations.
Another major accomplishment was implementing financial safeguards that distinguish supported financial effects from hypothetical interventions and identify unresolved funding needs.
We also developed visualizations that allow users to compare projected cash flow outcomes and better understand the trade-offs between financial decisions.
Finally, we're proud of our interdisciplinary collaboration. We combined financial economics, data analytics, software engineering, and artificial intelligence to create a practical financial decision-support platform.
What makes us especially proud is that Tangerine Mission Control doesn't pretend every financial problem has an immediate solution. Sometimes the most responsible recommendation is to preserve liquidity, identify the remaining funding gap, and clarify what information is needed before taking action.
What we learned
This project taught us that good financial decision-making requires more than accurate predictions or sophisticated AI models.
A recommendation can look mathematically attractive while being financially unrealistic if it relies on unavailable funds, unsupported assumptions, or obligations that cannot actually be changed.
We learned the importance of distinguishing financial indicators from real-world evidence and separating AI-assisted reasoning from deterministic financial calculations.
We also learned how important timing is. Two customers with similar monthly income and spending can experience very different financial pressures depending on when payments and deposits occur.
From a technical perspective, we gained experience building an application that integrates large financial datasets, AI services, financial modeling, validation systems, and an interactive user interface.
Most importantly, we learned that the value of financial technology isn't simply in generating more information. It's in making that information understandable, actionable, and trustworthy.
What's next for Tangerine Mission Control
Our long-term vision is to develop Tangerine Mission Control into a more comprehensive financial resilience platform that helps banks support customers proactively rather than waiting until financial difficulties become severe.
Our next priorities include expanding real-world scenario testing, improving the quality of personalized financial recommendations, and strengthening the system's ability to incorporate verified, up-to-date customer information.
We also want to explore more flexible savings strategies for customers with irregular income, strengthen payment timing analysis, and develop clearer explanations of financial trade-offs.
In a real banking environment, Tangerine Mission Control would require additional security, consent, privacy, suitability, and regulatory safeguards before supporting customer decisions.
Ultimately, we envision Tangerine Mission Control helping financial institutions shift from reactive financial assistance to proactive financial resilience planning, giving customers a clearer understanding of their options before financial uncertainty becomes a crisis.
