Wearable smart devices, which continuously capture physiological signals such as heart rate, respiratory patterns, and blood oxygen levels, offer significant potential for the early detection of serious health conditions. Timely diagnosis of diseases such as arrhythmia and sleep apnea can greatly enhance patient outcomes by enabling early intervention. However, extensive collection of diverse, real-world wearable sensor data faces challenges due to privacy concerns, data scarcity, and logistical constraints. This research introduces a novel deep learning framework that integrates publicly available wearable sensor data with synthetic physiological signals generated by large language models (LLMs) to create comprehensive and privacy-compliant hybrid datasets.The proposed framework leverages convolutional neural networks (CNNs), optimized for time-series data analysis, alongside advanced machine learning techniques to identify early signs of arrhythmia, sleep apnea, and related health conditions from physiological data. The integration of synthetic data generated by LLMs addresses critical challenges of limited data availability and privacy concerns, enriching the training datasets with diverse scenarios and physiological variations. Preliminary experimental results demonstrate that the hybrid approach, combining publicly accessible wearable sensor data and LLM-generated synthetic signals, significantly enhances the model's accuracy, generalization capability, and resilience to data variability. Models trained on hybrid datasets consistently outperform those relying solely on real-world data, suggesting that synthetic data provides meaningful supplementation to traditional datasets. This study further highlights how synthetic physiological data can enhance the scalability and efficacy of AI-based health monitoring systems, reducing dependency on extensive clinical data collection. By exploring and validating this innovative data synthesis approach, the research contributes significantly to developing more effective, accessible, and proactive healthcare diagnostic tools, ultimately advancing AI-driven solutions in preventive healthcare.

The iPELiNT project develops an AI-powered patent analysis dashboard designed to streamline the patent prosecution process for attorneys and practitioners. This web application leverages modern technologies including Vue.js with Nuxt3 framework for frontend development, NodeJS with Express for backend services, MongoDB for database management, and integrates AI models from OpenAI to analyze patent documents.

The system features a user-friendly dashboard that allows practitioners to upload patent applications, analyze document health, view CPC prediction analytics, examine keyword relevance, and identify potential prior art conflicts. Key functionality includes document parsing, automated health checks, Art Unit prediction, and generation of actionable reports. The solution also incorporates user account management, notification systems, and specialized document generation tools.

Development followed an iterative process with clearly defined milestones and tasks distributed across team members. The project addresses a critical need in the patent industry by providing an all-in-one platform that simplifies complex patent analysis, replacing traditionally fragmented and cumbersome tools with a streamlined, intuitive interface.

The completed iPELiNT dashboard enhances efficiency for patent professionals, improving application quality through AI-powered insights, and ultimately streamlining the patent prosecution workflow with modern design principles and cutting-edge technology.

Designed to empower students with transparent, real-time insights, this innovative digital platform provides comprehensive reviews of classes and instructors, enabling informed academic decision-making. It aggregates detailed evaluations of course content, teaching effectiveness, workload, and overall classroom experience, offering a dynamic alternative to traditional end-of-semester surveys that frequently deliver delayed or insufficient feedback. Backed by survey research underscoring the vital role of timely, honest assessments in shaping successful academic journeys, the platform bridges the gap between institutional data and the practical needs of students. Its intuitive, user-friendly interface allows seamless navigation through a wealth of peer-generated feedback, making it easier for students to select courses that align with their educational goals and personal learning styles. Moreover, by establishing a constructive feedback loop, it provides educators with actionable insights to refine their teaching methods and foster continuous improvement. This collaborative environment not only enhances individual learning experiences but also contributes to building a more effective, accountable educational community. Through open dialogue and shared knowledge, the platform drives positive change, promoting excellence and ensuring that every academic decision is supported by reliable, student-centered information. By continuously evolving based on extensive user feedback, the platform remains dedicated to advancing educational quality and student success.

Public property tax data is often presented in raw formats, making it difficult for the average user to interpret. Our client initially developed a product that provided access to Kern County property tax information only. Our project enhances accessibility by developing ParcelSearch.com, a platform that centralizes property tax data. With this rebranded system, we have expanded coverage to include Kern, Monterey, and Tulare Counties, with plans for further expansion. Users can create accounts and choose from various subscription plans to conduct property searches using multiple search criteria: owner name, parcel number, and legal descriptions. With the development of a user-friendly interface and expanded search functionalities, the platform caters to realtors, investors, and homeowners seeking property insights. This system was built using modern web technologies, including Vue.js for the frontend, Java and Spring Boot for the backend, and PostgreSQL for database management, to name a few. Future plans include expanding nationwide to create an all-encompassing and user-friendly property data platform.

Longspeeds is an innovative e-commerce platform designed to streamline the buying and selling of auto parts, providing a seamless experience for both individual customers and automotive businesses. The platform offers a comprehensive catalog of high-quality parts for a wide range of vehicles, from everyday cars to performance and specialty models. Leveraging modern web technologies such as Next.js, Longspeeds ensures fast performance, responsive design, and secure transactions. Key features include advanced search and filtering, user-friendly inventory management, real-time order tracking, and support for both retail and wholesale transactions. With a focus on reliability, affordability, and user satisfaction, Longspeeds aims to become a trusted destination for auto enthusiasts and mechanics alike.

Academic advising presents significant challenges in both time management and complexity. Currently, students navigate between two advising options: generic online resources and personalized consultations with professors and advisors. While personalized advisement offers tailored advice, professors cannot be expected to meet with every undergraduate in their major, especially as enrollment grows, and academic advisors may lack specialized knowledge required for some majors. Echelon addresses this gap by creating a middle ground between generic and personalized advising, offering students an effective supplement and saving time for all parties involved. Echelon functions as an intelligent chatbot assistant powered by large language models such as Llama 3 and Mistral. Upon signup, students share their academic records, enabling Echelon to create individualized profiles that consider key factors such as major/minor selection and performance in critical courses. The project is being built using TypeScript and Rust with Svelte and Axum frameworks, respectively. Echelon utilizes PostgreSQL for user account and conversation storage and Qdrant for vector storage and retrieval. Designed with flexibility in mind, Echelon can be deployed at any university, given basic institutional information such as course catalogs and degree requirements.

Cryptococcus is an invasive fungus that causes cryptococcosis, an infection that highly affects immunocompromised people. There are currently a limited number of antifungals available to treat Cryptococcus, and with the increased in antimicrobial resistance, we need different alternatives to treat fungal infections. Our lab has identified proteins involved in the synthesis of eicosanoids, which are lipid signaling molecules involved in regulating the immune response. Moreover, fungi can produce eicosanoids using different enzymes that humans do, opening a line to identify new drug targets using these pathways. Previously, our lab had identified genes upregulated in the presence of the eicosanoid’s precursor, arachidonic acid. Our goal is to use bioinformatics to predict and characterize the protein structure, using AlphaFold2, a machine learning application based on a deep neural network, and using this tool, identify small molecules that will bind to the proteins and help make drug design more efficient.

FWDX BioBlade is a web-based system designed to automate genetic sequence comparison for Fort Worth Diagnostics (FWDX), a company specializing in high-quality diagnostic reagents. FWDX faces a significant challenge: ensuring its reagents remain effective as pathogens mutate over time. Currently, this process is manual, time-intensive, and costly, relying on external bioinformatic consultants to compare existing reagent sequences against national and international genetic databases like NCBI and GISAID.

BioBlade improves this workflow by automating sequence comparisons, detecting mutations or deletions, and generating real-time reports. This automation significantly reduces turnaround time, improves accuracy, and lowers costs, empowering FWDX scientists and regulatory personnel with timely and accurate information. Key features include:
- Automated sequence analysis for efficient reagent validation
- Customizable query intervals for up-to-date comparisons
- A user-friendly reporting dashboard for streamlined decision-making

As artificial intelligence and machine learning continue to evolve, the need for efficient search and retrieval mechanisms for unstructured data has grown exponentially. Traditional relational databases, optimized for structured queries, struggle with the high-dimensional nature of modern AI-generated embeddings. This challenge has led to the rise of vector databases, specialized systems designed to store, index, and retrieve data based on similarity rather than exact matching.

This symposium explores the fundamental concepts of vector databases, their key components—such as vector embeddings, indexing techniques, and similarity search algorithms—and their advantages over traditional databases. We discuss how vector search operates using distance metrics like cosine similarity and Euclidean distance and compare the roles of vector databases and standalone vector indexes.

This project presents an interactive forecasting tool for Influenza-like Illness (ILI) trends using historical CDC data and machine learning models. Built with Python and Streamlit, the app enables users to visualize yearly ILI patterns, compare predictive models, and forecast future cases based on recent trends. Three models — Linear Regression, Random Forest, and XGBoost — were evaluated using Root Mean Squared Error (RMSE) and R² Score. Surprisingly, Linear Regression achieved the best performance with an RMSE of 0.106 and R² of 0.960, indicating that simple models can be effective for this type of time-series data. The app also includes features for dynamic forecasting and CSV export, making it a practical tool for public health analysis and planning.