La Jolla, CA · Cumulative GPA: 3.6

View credential PDF

Triton AI Member & Autonomous Racing Team · ROS, Linux, Python, Robotics, Simulation

• Advance applied AI, ML, robotics, and autonomous systems initiatives through Triton AI and the UCSD Autonomous Racing Team.
• Contribute to technical discussions, workshops, and project development across AI/ML, LLM workflows, ROS, and simulation environments.
• Supported a 2nd-place finish at a Purdue University competition by documenting 3+ technical subsystems of a ROS/Linux autonomous vehicle project.

Real Estate Data Science Platform · Remote, USA

• Engineered a client-facing ML pipeline using XGBoost, Random Forest, and Linear Regression integrated through Cotality's Trestle API.
• Achieved R² = 0.89 and about 6% MDAPE across 5+ home-price segments, then translated results into executive decision briefs.
• Reduced analysis turnaround time by 40% through a standardized EDA, feature engineering, tuning, and evaluation workflow across 5+ experiments.

Supply Chain Operations · Jakarta, Indonesia

• Structured a stock-out reduction problem into a forecasting project and built ARIMA and Decision Tree models on 9 months of procurement data.
• Improved forecast accuracy by 15% and reduced stock-outs by 20% for purchasing stakeholders.
• Built automated data ingestion pipelines for Tableau dashboards across 8+ procurement and inventory metrics, saving 3 hours per week.

Printing supplier for Indonesian hospitals and clinics · Hybrid/Remote

• Identified hidden over-ordering patterns through time-series analysis, supporting a 15% reduction in procurement costs.
• Designed automated Excel-based expense-tracking and budget-planning workflows, recovering 4 hours per week of manual reconciliation time.

Python, pandas, Streamlit, Claude API, Pricing Analytics

• Built a schema-adaptive analytics application that standardizes restaurant POS data and computes revenue, contribution margin, popularity, menu quadrants, demand elasticity, and pricing scenarios.
• Developed configurable price simulations and item-level recommendations, with validation that prevents unsupported profit analysis when cost data is unavailable.
• Integrated Claude to translate deterministic Python analytics into management-ready recommendations while keeping financial calculations independent of the LLM.

Python, EEG, MOABB, Signal Processing, SVM

• Built an EEG biometric authentication pipeline using resting-state recordings from 40 subjects and 310-dimensional bandpower features extracted across 62 EEG channels.
• Achieved mean validation AUC of 0.989 and evaluated both within-session and cross-session verification to quantify biometric performance across days.

Python, PyTorch, EfficientNet, Computer Vision

• Fine-tuned EfficientNet-B0 to classify Southern California locations across six geographic classes using transfer learning and data augmentation.
• Reached 91.8% validation accuracy through staged training and test-time augmentation.

Python, scikit-learn, pandas, Statistical Modeling

• Built OLS, Ridge, and Lasso regression models using CDC PLACES data across 2,299 U.S. counties and seven health and social-need predictors.
• Achieved R² = 0.628 with 5-fold cross-validation and evaluated model performance, feature relationships, and regularization effects.