Mathias Villerabel

Built an IFRS 17 actuarial/financial simulation engine (cash flows, discounting, Contractual Service Margin) to benchmark EY outputs and ensure regulatory compliance. Automated end-to-end reconciliation (GL ↔ actuarial models), cutting manual work from multi-day reviews to < 2h/run and improving auditability & reproducibility. Designed and optimized PySpark pipelines in Palantir Foundry, scaling to 109+ accounting records per batch and integrating with downstream financial reporting systems.

Developed time-series forecasting and ML models (ARIMA, gradient boosting) and designed a custom PyTorch framework for scalable training / inference. Improved 3-month demand forecasts by 12% vs. FA baseline.

Automated detection of new construction from SAR time series (InSAR coherence + intensity). Delivered a cloud-based Earth observation platform with dynamic caching on Google Cloud, reducing data latency for SDG indicators in economics and environment. Built scalable object detection models for maritime trade (containers, cars, trucks, ships); awarded 2nd Prize at the NEDO Challenge. Applied Earth observation and machine learning methods during NASA, ESA & JAXA hackathons to address real-world challenges such as wildfire monitoring and prediction.