Orionx (Crypto Exchange) · Santiago, Chile

• ▸Designed and deployed an automated trading system in Python on Kubernetes, scaling administered capital 4x through optimized execution and message queue architecture.
• ▸Engineered 100+ ETL pipelines (Airflow, PySpark) for real-time market data processing, automating discovery of 30+ arbitrage opportunities and reducing data processing time by 80%.
• ▸Built a convex optimization model (CVXPY) for real-time asset allocation, increasing capital utilization from 40% to 95% across the portfolio.
• ▸Implemented an online learning algorithm for dynamic bid/ask estimation, reducing slippage risk by 50% during volatile market conditions.
• ▸Developed triangular arbitrage execution using Bellman-Ford graph traversal, generating $100K/month additional volume at a 67% win rate.
• ▸Shipped mission-critical APIs (FastAPI, Pydantic, SQLModel) that enabled a 3x increase in daily trade volume with full data integrity guarantees.
• ▸Established production-grade CI/CD with MyPy, Pytest, and Pydantic validation, significantly reducing deployment errors.

Agrosuper · Chile

• ▸Developed and implemented a random forest ensemble model in Python to predict risk factors in slaughterhouse workers from inertial sensor data, resulting in a classifier with 98% out-of-sample accuracy.
• ▸Measured intervention effectiveness by designing an A/B test experiment, leading to a 30% reduction in worker risk exposure and an 8% reduction in operational time.
• ▸Worked closely with stakeholders to communicate data findings, driving data-driven decision-making across departments.

Institute for Mathematical Engineering UC · Santiago, Chile

• ▸Developed regression models to forecast agricultural yields, reducing error rates from 20% to 2% and enabling more accurate financial planning.
• ▸Created detailed data visualizations and reports to communicate insights to stakeholders, facilitating data-driven decision-making.

Chilean Central Bank · Santiago, Chile

• ▸Conducted statistical analysis on Chilean household debt data, leading to the identication of the top three factors in debt underreporting.