- Led the development of an automated arbitrage system in Python, doubling administered capital from $250k to $500k.- Developed and maintained CI/CD pipelines on GitHub Actions, ensuring 24/7 availability and performance of tradingsystems. - Leveraged AWS (EC2, S3, Lambda, EKS, ECR) to enhance the scalability and reliability of the automated trading system.- Built and optimized ETL pipelines using Airow and PySpark for market data processing and analysis, discovering 30+ under-arbitraged assets, which increased monthly volume from $200k to $1M.- Implemented a convex optimization model using CVX in Python for real-time asset allocation, significantly improving asset utilization from 40% to 95%.- Collaborated closely with cross-functional teams, including OTC, compliance, marketing and engineering, to successfully ship three new APIs for external use using Fast API.- Implemented an online learning algorithm to derive robust bid/ask price estimates from volatile order book data, resulting in a 70% reduction in slippage risk.- Created and optimized data pipelines using Airow, MongoDB, PostgreSQL, and S3 to automate reporting and ensure compliance, creating Metabase dashboards displaying over 20 financial KPIs for decision-making.

Lead data scientist in a joint venture between DICTUC UC and Agrosuper S.A.I developed with full ownership a prototype for the detection of musculoskeletal disorders among slaughterhouse workers by using information from low-cost sensors located on the wrists. The objective was to determine the feasibility of modeling the presence of risk factors during meat cutting activities as a time-series classification problem, and use such technology for decision making.My engineering tasks included:(1) Hardware to measure worker movements using low-cost sensors.(2) ETL: Ingest sensor data (via bluetooth protocol) into S3 data lake, transform data (time indexing, labeling, etc) into RDS-SQL warehouse. (3) Supervision: User interface to visualize and label the presence of risk factors.(4) Feature Engineering: Used expert ergonomic insight and research to engineer novel features that capture the presence of relevant risk factors. (5) Negotiate success metrics, and defined statistical tests to evidence the impact of the prototype, with senior executives.(6) Implement a pipeline to train and predict of the presence of risk factors over movement data: pre-processing, feature engineering, feature selection, training, validation, model selection and inference. Developed using Python, Sklearn, AWS EC2 and PySpark.(7) Generate decision rules to react across risk levels.As a professional :(1) I thought deeply about a complex data science problem.(2) Engage and negotiate with senior executives and end users(3) Learn new technologies and frameworks to deal with real-world dataAs a conclusion of the project:(1) Demonstrated the potential of low-cost sensing technologies in a data-driven framework, to produce assess the presence of risk factors.(2) Substantial economic value in terms of risk management and productivity can be gained from the preventive detection of risk factors.

Responsible for translating complex mathematical and statistical concepts into useful insights for non-technical , business-oriented students with varying roles (junior to senior) in Latin American companies.

Worked as a consultant for the IMC-UC.My role was to: (1) translate the business requirements of the client into problems that can be addressed with statistical learning tools, (2) implement solutions using state of the art techniques, (3) find valuable insights in the data. The core of the project was to use historical data to generate accurate forecasts for critical indicators of the daily operation, reducing planning error. The condition for success was to improve on current human forecasts.My activities involved:(1) Engagement with the client to understand the business problem and the data. I negotiated and proposed the set of metrics to determine the success of the project.(2) Build an ETL pipeline to enable analytics. Raw data was available in spread sheets scattered across the organization. I designed a Python program to collect and parse inputs, and then store/organize the transformed data into SQL tables.(3) Build a time-series forecasting pipeline in Python and Sklearn. I developed a methodology based on state-of-the-art research in time-series prediction, considering variable window size (input) and variable step size (output). Feature engineering, selection and dimensionality reduction were also added. We trained an ensemble of weighted Random Forest and LightGBM regressors, and validated the models by back-testing the performance of the algorithms across time. Hyper-parameters were fitted through cross-validation.(4) Communicate/ report the results and insights obtained by using visualization tools such as Tableau, Matplotlib and Plotly.As a result of the project:(1) We achieved a MAPE error below 5%, improving on current human performance (MAPE of 10% on average). (2) We identified key features that influence the target operational variables, providing hypothesis for further operational improvement.(3) We raised gaps in the sensing capabilities needed to generate production-level implementation.

- Co-organizer of the First National Mathematical Engineering Encounter (2015).- Student Chapter Staff (2015 - 2018)- President (2016).

Intern at "Gerencia de Investigación Financiera". I worked in projects related to the Household Financial Survey (EFH), focusing household debt modeling. I was expected to quickly get acquainted with state-of-the-art research on household finance, and to effectively communicate my findings to senior economists, prioritizing insights related to monetary policy.During my time at the Central Bank of Chile, I:- Implemented the Gale-Shapley algorithm in R to match official debt records (from the SBIF) with the self-reported debt registers (from the EFH). - Implemented a variety of generalized linear models in R and STATA to associate debt under-reporting by chilean households with debt-type and demographic information.

I was a teaching assistant of the following courses:- MAT1610 Calculus I (Civil Engineering). - Mat 1630 Calculus III (Civil Engineering). - MAT210E Calculus I (Economics).- MAT220E Calculus II (Economics). - ICS3153 Advanced Optimization (Industrial Engineering). Mainly theory and algorithms of convex optimization to real life applications. Among other topics, the course covers general convex optimization theory, first order algorithms, SDP programing and interior point methods. - IMT3500 Mathematical Engineering Capstone (Mathematical Engineering). Project course in collaboration with companies. The aim of the course is to solve a real industry problem using the tools and capabilities developed during the mathematical and industrial engineering program. The projects generally relate to finance, operations managment and business analytics. The problems are usually solved using data science tools and numerical simulations.

Worker practice in the "Reposición" area at Homy Store, Parque Arauco.