Biokind Analytics at UC San Diego (Non-Profit Organization)
Biokind Analytics at UC San Diego is a student-led organization that partners with local healthcare and social-impact nonprofits to solve real-world operational challenges through data science. Our teams develop bespoke, data-driven solutions using exploratory and predictive analytics, AI and large language models, cloud computing, APIs, and large-scale data processing.
Through these projects, students apply their academic training to meaningful problems while helping clients better understand their programs, donors, grants, operations, and investment decisions. Our work emphasizes practical, intuitive, and actionable results, delivered through dashboards, web applications, and presentations, with more than $140 million in combined financial analyses performed for our partners.
Digital Bouncers: Secure Smart Home Solution with Large Language Models (Data Science Graduation Capstone Project)
As AI-driven smart home assistants become more prevalent, ensuring their security and reliability is essential. Digital Bouncers is an LLM-powered chatbot designed to provide homeowners with safe and accurate insights into their energy usage while mitigating AI-related risks. By integrating a Text-2-SQL LangChain agent, a LangGraph persistence layer, and cloud data storage in PostgreSQL, this solution enables dynamic, structured data retrieval and real-time contextual interactions. Advanced guardrails at the input, intermediary, and output levels ensure accuracy, prevent hallucinations, and protect users from misleading or harmful interactions. This project demonstrates an innovative approach to AI safety in smart home environments, setting a new standard for secure and intelligent home automation.
Statistical Inference for U.S. Presidential Elections (Team Project)
A detailed analysis of twelve elections for the U.S. presidency between 1976 and 2020 for a deeper understanding of their nature; analyses performed include exploratory, goodness of fit (LRTs), similarity/dissimilarity analysis of distributions, predictive (ML), plus a fun interlude mingling of Bayesian and classical (frequentist) statistics.
OPUBOD - Occupancy Prediction Using Building Operation Data (Personal Project)
A time-series machine learning project to discover useful building operation metrics for predicting occupant count; examined models include linear regression, auto-regression, and recurrent neural nets, plus statistical analyses for feature selection.
Developed game recommender systems for (1) play prediction using collaborative filtering and (2) hours played prediction using latent-factor models with coordinate descent and gradient descent. All models implemented from scratch with analogous performance to those of the Surprise library from peers following a class leaderboard.
Solar-Powered Smart Lighting for Sustainable Living (Team Project)
Built a solar-powered data science IoT device capable of collecting live sensor data, adjusting lighting depending on environmental factors for energy efficiency, and sending data to a MySQL database using MCU on-board Wifi (synchronous communication).
Performed an analysis of solar charging voltage under different scenarios and developed a data processing script to query/transform/analyze/visualize data.
Exploratory and Predictive Analytics for Precision Medicine (Team Project)
A data science project for Personalized and Secure Drug Discovery supported by the Semiconductor Research Corp. Scholars Program and PRISM Center at UCSD, mentored by Dr. Niema Moshiri. The project combines intensive exploratory and predictive analyses to identify patterns and predict breast cancer recurrence using biomarkers, medical, and lifestyle data from over 3,000 patients.
A comprehensive look into major power outages in the U.S. to assess statistical relationships (part 1) through exploratory analysis, hypothesis testing, data missingness assessment, and predictive power (part 2) using supervised machine learning models and feature engineering, built in pipelines with evaluation comprising diverse accuracy metrics and cogent assessments.
Conducted a comprehensive analysis of global happiness patterns using public data from the 2022 World Happiness Report to uncover insights into the factors influencing well-being worldwide.