Ginseng

Landing Page
Registration Page
Demo Page
Financial Data

Inspiration

People with irregular income often have difficulty planning how much cash they need for the near future. We wanted to build a system that uses historical financial behavior and uncertainty to help users plan their liquidity for the next 30 days.

What it does

Ginseng is a probabilistic liquidity planning system for variable-income investors.

Simulates 5,000 possible 30-day financial futures
Calculates a 95% liquidity reserve
Identifies potential cash shortfalls and their severity
Compares funding strategies such as credit, investment liquidation, and spending cuts
Uses CVaR optimization to find a lower-cost funding strategy
Uses Gemini to explain the quantitative results in plain language
Integrates Capital One Nessie for financial data

How we built it

We built Ginseng as an end-to-end pipeline: Financial data → Simulation → Risk analysis → Optimization → Recommendation

Historical daily financial data is processed into time-series features
Stationary block bootstrap generates 5,000 possible 30-day scenarios while preserving temporal patterns
Risk metrics measure cash exhaustion, shortfall severity, and time below the liquidity buffer
CVXPY and the Rockafellar–Uryasev CVaR formulation optimize funding decisions
Gemini API explains the model's results to users instead of replacing the underlying quantitative model
Capital One Nessie provides sandbox financial data such as accounts, transactions, merchants, and bills

Challenges we ran into

Modeling irregular and unpredictable income
Preserving time-series patterns in simulated financial futures
Turning uncertainty into actionable funding decisions
Balancing liquidity needs with investment assets
Making quantitative results understandable to non-technical users
Integrating multiple APIs and systems into one workflow

Accomplishments that we're proud of

Built an end-to-end financial simulation and optimization system
Generated 5,000 simulated futures for each planning scenario
Implemented a 95% liquidity reserve framework
Applied CVaR optimization to funding decisions
Compared multiple liquidity strategies instead of producing a single estimate
Connected quantitative outputs to a Gemini-powered explanation layer

What we learned

Financial planning needs to account for uncertainty, not just averages
The severity and cost of a financial shortfall matter as much as its probability
CVaR can connect simulated risk to practical decisions
Time-series structure is important when generating financial scenarios
AI is most useful when it explains and communicates real model outputs