LISI Automotive, Madrid, Spain January 2025
- Collaborated with LISI Automotive in Madrid on an international consulting project to address a real-world business problem.
- Fine-tuned and customized pre-existing forecasting models (Holt-Winters, XGBoost) using hyperparameters, lag features, and rolling averages to meet client-specific needs., Reduced MAPE from 20% to 7.49%, enabling tailored solutions for accurate sales and raw material forecasts.
- Trained models on 80% historical sales data (2019-2023) and validated using 2024 actuals, improving production and inventory planning.
DSA Alliance NPO, San Diego, CA, December 2024
- Analyzed shelter utilization trends in San Diego to support homelessness reduction strategies.
- Merged HUD and economic datasets to uncover key correlations, finding that a $100 rent increase results in 322 additional homeless individuals, while a $1,000 per capita income rise reduces homelessness by 283.
- Identified emergency shelters as the most effective, reducing unsheltered rates by 20.13 individuals per unit increase in utilization.
- Proposed expanding HMIS programs, increasing shelter capacity, and refining policies to improve resource allocation and operational efficiency.
Spotify Recommender System, San Diego, CA, November 2024
- Designed a playlist recommendation system using the Spotify Million Playlist Dataset (MPD) with advanced machine learning techniques.
- Applied Word2Vec for embedding generation and BERT4Rec for sequence modeling, achieving 96.13% accuracy in playlist continuation.
- Enhanced user engagement and music discovery by integrating metadata, audio features, and behavioral insights into the model.