In this page you'll find the extended version of my resume (all projects, 'Clubs & Leadership' and past education) or feel free to just download the 1-page version!

Main Resume

University of California, Berkeley (Berkeley, California)

B.A. Data Science with Domain Emphasis in Industrial Analytics (CDSS college)
Relevant Coursework: Data Engineering, Data Mining & Analytics, Probability for Data Science, Data Inference & Decisions, Principles & Techniques of Data Science, Data Structures & Algorithms.
Awards & Scholarships: Generation Change Scholar, James Hjul Scholar, Allmond Scholar, Albert Job Scholar.

IB diploma at Centro Escolar el Roble (Guatemala City, Guatemala)

Summa Cum Laude
Activities and Societies: El Roble soccer team, volunteering at Guatemala's neurological institute (Instituto Neurológico de Guatemala) and the American Red Cross.

Data Science Intern

University of California Berkeley, Financial Aid Office (FAO)

Berkeley, California

– Cleaned and integrated multi-year student and aid datasets into relational schemas, optimizing SQL queries for large-scale reporting.
– Ran A/B tests on outreach emails and student portals to boost FAFSA completion and critical financial-aid form response rates.
– Built AWS/GCP dashboards on aid usage, unmet need, retention, and first-gen/low-income outcomes to support

Data Science Intern

U.S. Department of Health and Human Services (HHS)

Washington DC (Remote)

– Developed regression and optimization models and A/B tests on CIL funding and utilization data (2019 vs. 2022) to quantify post-pandemic underperformance.
– Helped to automate ETL pipelines using Python and APIs, reducing data processing time by 30%.
– Built BigQuery and Tableau dashboards summarizing funding and usage trends for non-technical decision-makers.
– Synthesized pre-pandemic and post-pandemic funding and utilization findings with executive directors, contributing statistical analyses and visualizations to a federal report scheduled for release in 2026.

Lead Researcher @ CDSS Data Discovery Program

UC Berkeley College of Computing, Data Science, and Society

Washington DC (Remote)

-Led a research team for the U.S. Health Department branch (ACL), designing and implementing advanced ETL pipelines to clean and structure unorganized datasets for 350 centers nationwide.
-Developed SQL and PostgreSQL-based data models, optimizing storage solutions to ensure efficient query performance and scalability.
-Engineered comprehensive end-to-end workflows to extract, transform, and load (ETL) data, providing actionable insights that directly influenced federal policy decisions.

Endorsements & Outcomes: Causal & Predictive Analysis of 2022 U.S. Primaries

-Asked two things: do endorsements cause wins, and can we predict results?
-Cleaned/merged FiveThirtyEight data; engineered features (endorsements, incumbency, fundraising, state/party).
-Causal: drew a DAG to pick confounders; estimated propensity scores (logit); nearest-neighbor matching + IPW; checked balance (SMD); estimated ATE/ATT with bootstrap CIs.
-Predictive: stratified splits; L1/L2 logistic regression and class-weighted Random Forest; threshold tuning.
-Validation: 5-fold CV, ROC-AUC/precision/recall/F1, calibration; model explanation via permutation importance/SHAP.
-Result: endorsements—especially Trump’s—and incumbency were most predictive; best RF reached ~0.78 F1.

Streaming Analysis(ML)

-Engineered a churn label from Netflix user subscription data using join/last-payment dates and computed overall churn rates. -Built churn prediction models in Python with pandas and scikit-learn, including MLP neural networks, decision trees, and tuned random forests with GridSearchCV and feature importance. -Reduced feature space with PCA and performed customer segmentation using K-Means clustering. -Used IsolationForest to flag high-risk churn users and visualized churn patterns by country, age, device, and cluster with matplotlib/seaborn.

Yelp Insights and NoSQL Data Processing with MongoDB

-Modeled Yelp users, businesses, and reviews across MongoDB and PostgreSQL to compare document vs relational schemas and join strategies. -Wrote MongoDB aggregation pipelines (via PyMongo) for geospatial filters, city/state rollups, and keyword-based “to_avoid” review flags using text search. -Built SQL queries and joins in PostgreSQL to mirror MongoDB lookups and analyze performance differences with EXPLAIN ANALYZE. -Sampled data into pandas to study missingness, explode categories, and construct cleaner business feature tables for downstream analysis.

Campus Sensor Data Cleaning and Time-Series Interpolation in PostgreSQL

-Cleaned and standardized campus HVAC and energy sensor data in PostgreSQL, resolving inconsistent units and messy building/location metadata. -Used SQL (including JSON extraction) to explore schema, validate key relationships, and produce analysis-ready sensor tables. -Implemented robust outlier detection and winsorization with median/MAD to stabilize noisy sensor readings. -Built regular 15-minute time-series grids using GENERATE_SERIES and window functions with forward/backward fills and linear interpolation for missing data.

Neural Network for Survival Rates in Titanic

-Built an end-to-end survival prediction model on the Titanic Kaggle dataset. -Engineered features and handled missing data. -Used scikit-learn pipelines for preprocessing plus a regularized neural network with dropout and batch norm. -Evaluated model performance and generated a Kaggle-ready CSV submission of survival predictions.

BYOW (build your own world)

-Generated a 2D avatar world based on different data structures (using a user's seed so worlds are repeatable and deterministic) by writing +3000 lines of Java quality code.
-Guaranteed a connected map and smooth avatar movement; used BFS for connectivity and a wall/floor collision map with W/A/S/D loop.
-Implemented save/load and testing; serialized game state (:Q/L), parsed inputs, and wrote a range of JUnit tests.
-Employed LinkedLists, ArrayLists and HashSets to provide rapid data retrieval and management.

NGordNet

-Browser-based tool for exploring the history of word usage in English texts using large datasets (50,000 words).
-Ingested Google N-grams and WordNet; implemented TimeSeries/NGramMap with HashMap/TreeMap, normalized frequencies, and efficient year-range aggregation.
-Modeled WordNet as a synset DAG; computed hyponyms via BFS/DFS + set unions; cached results to avoid repeated traversals.
-Exposed endpoints with lightweight Java web handlers (Spark/Jetty scaffold), returning JSON for frequency-filtered hyponyms across time windows.
-Wrote JUnit tests; profiled asymptotics; reduced I/O via memoization and precomputed indices.

Spam and 'Ham'

-Loaded and cleaned labeled spam/ham emails in pandas (lowercasing, missing-text imputation). -Explored keyword patterns in spam vs ham using custom binary word indicators and seaborn visualizations. -Engineered a feature matrix with a words_in_texts function capturing presence of selected terms. -Trained and evaluated a scikit-learn logistic regression spam classifier vs a naive baseline using accuracy, precision, recall, and false-positive rate.

Computer Science Intern

Open Project @ Berkeley

Berkeley, California

-Participated in one of Berkeley's biggest CS clubs.
-Aimed to have a diverse impact on a plethora of student problems. In this semester my team and I were working to make a better version of the University's official portal for students.

Forum Committee Vice President

Latin American Leadership Society @ Berkeley

Berkeley, California

-Organized and executed large-scale events for 100+ attendees, leveraging data-driven tools to streamline scheduling and participation.
-Utilized technology platforms to optimize communication and solve logistical challenges, enhancing overall event execution.
-Promoted awareness of complex societal issues affecting Latin American communities both in the U.S. and abroad, fostering meaningful discussions and solutions.