All Case Studies
Financial ServicesSeries B Fintech2025
Atlas Forecasting Engine
Replaced a brittle rules-based forecasting system with a hierarchical time-series model that improved cash-flow accuracy by 38% and unlocked treasury automation.
Client
Series B Fintech
Sector
Financial Services
Duration
14 weeks
Team
3 engineers · 1 researcher
The Challenge
The starting point.
The client's treasury team relied on a 12-year-old spreadsheet model to forecast working capital across 40 currencies. Forecasts drifted weekly and required constant manual reconciliation, blocking automated FX hedging.
The Approach
What we built.
- 01Audited two years of historical cash positions and identified seven seasonality regimes
- 02Built a hierarchical Prophet + LightGBM ensemble with currency-aware features
- 03Designed a continuous-evaluation harness running against live treasury actuals
- 04Shipped a FastAPI service with an auditable forecast lineage for compliance
Results
What shipped, what changed.
38%
Forecast accuracy lift
$4.2M
Annual hedging savings
92%
Manual reconciliation removed
14 wk
From kickoff to production
Stack
What's running in production.
PythonProphetLightGBMSnowflakeFastAPIAirflow
Next case study
Aurora Vision Platform