Department of Computer Science

Francisco Soltero, Raiya Dhalwala, Ritika Mishra, Zoë Dupre

Our project, LastCall, is a bar-specific scheduling app designed to streamline scheduling challenges and boost productivity. It features a user-friendly interface with distinct views for employees and managers. Employees can log their availability and access their weekly schedules, while managers can efficiently create, review, and publish schedules. Key features include caution flags for double shifts or back-to-back closing-to-opening transitions, as well as customizable scheduling options for different roles such as bartenders, barbacks, and bouncers. While competitors like 7shifts cater primarily to restaurants, they lack the flexibility needed for bars and other niche hospitality businesses. LastCall is specifically tailored to the dynamic nature of bar operations, providing a centralized, adaptable scheduling solution to meet their unique need

Gabe Epstein, Jared Markowitz, Kayla Willis

Our project, in development for the NASA RASC-AL Sustainable Lunar Evolution challenge, addresses the crucial challenge of ensuring astronaut safety from environmental and operational hazards on the moon. The lunar environment presents unpredictable risks such as solar flares, moonquakes, and operational failures like oxygen depletion, which endanger the safety of astronauts and the mission itself. We are designing the Lunar Advanced Warning and Risk Detection System (Lunar AWARDS) to provide real-time monitoring and early warnings for these hazards. This will take the form of a dashboard with the ability to display real-time data, simulate hazard alerts, and provide astronauts with a comprehensive view of potential risks while also supplying additional information on separate web pages.

Sam DeMarinis, Zoe Oboler, Miranda Diaz

This project presents the design and development of an innovative language learning application for children aged 3 to 7 years old. The app uses an open-ended, accessible, and immersive approach to language acquisition. This game will address the lack of open-ended language learning games for kids, as well as the limited multilingualism in the United States. The initial version of the app will be implemented with the target language of Spanish, since it is the second most-spoken language in the country. As opposed to following a strict curriculum, the game serves as a ‘virtual dollhouse’ in which the player can explore different settings and learn new vocabulary. A series of knowledge assessments will ensure that the user is effectively learning new words and phrases, individualizing the experience for each user. Additionally, animations and artwork will be engaging for young children, encouraging them to continue exploring and learning new words.

Izzy Blair, Hubert Mendez, Ethan Schaferkotter

Closed captions are vital for deaf and hard of hearing people, yet traditional models often lack tone, emotion, and visual differentiation, making them feel “dull” and “emotionless.” Building on Dr. Hassan's research on Rich Caption systems, we aim to create a musically focused closed-captioning system that enhances emotional connection to music through richer and more expressive visualizations. Our project will build upon previous versions of a Rich Caption Editor to add capabilities for visualization of different aspects of captions as well as music—specifically, we wish to add a PNG visualizer that can be animated upon as well as implement caption block color controls. The PNG visualizer will consist of a PNG image that the user can choose that will appear beside the captions whenever music is played in the video. We will also implement some animated features into the PNG visualizer that will reflect certain aspects of the video in real time—for example, the opacity of the PNG image will reflect the loudness.

Killian Daly, Duke Glenn, Henry Miller, Reid Miller

To many, a day in court facing a legal trial can be the worst day of their life, and a significant financial burden for anyone to take on. Our goal is to make the legal process more understandable for the average person by creating a user-friendly website that provides financial and judicial outcomes and connect users to proper legal next steps. We achieve this through the development of a machine-learning model that matches user input to likely outcomes based on similar cases from the past. Currently designed for vehicular crash claims cases, the website contains an interface which allows users to input their situation and key facts of the accident for our machine learning model to analyze and compare to its training data of hundreds of relevant court opinions and fact patterns. The website will respond with real time insights that help users quickly understand expected outcomes and connect them to proper legal representation that specializes in their case and area.

Peter Sapountzis, Bryan Flanagan, Griffin Olimpio, Rhon Farber, Jack Zemke

In collaboration with the Entergy AI Team, and with guidance from Andy Quick (SVP of AI) and Sean Douglass (AI Product Manager), our team worked with the Investor Relations and Compliance divisions to identify high-impact opportunities for AI to streamline their workflows. The key challenge we address is streamlining the time-consuming task of analyzing dense, long-form Public Service Commission (PSC) meeting videos. REGGIE (REGulatory Governance Inquiry Engine), a Retrieval-Augmented Generation chatbot tackles this challenge by providing an intuitive frontend chatbot to the teams. Using a combination of the YouTube API and Open AI’s Whisper model, we ingested and processed over 200 hours of PSC meeting footage, developing a retrieval system powered by the Weaviate vector database and Cohere embeddings. Streamlit powers the frontend chat interface, while FastAPI serves as the backend framework connecting the embedding model, vector database, and Anthropic’s Haiku model, which generates responses based on the retrieved transcript content. This setup allows users to query transcripts about regulatory decisions, rate cases, hurricane preparedness, and other utility-specific topics discussed in PSC meetings across multiple states. The system processes transcripts and timestamps, allowing users to filter searches by jurisdiction (Louisiana, Texas, Mississippi, Arkansas, and New Orleans) and retrieve exact answers with direct links to relevant video segments. With an easy to use chat interface and a powerful search functionality, REGGIE empowers Entergy teams to productively navigate PSC content with ease and precision.

Ella Moses, Claire Porier, Tyler Simms, Charles O'Bert

Court Watch NOLA is a local non-profit that promotes transparency and equity within the New Orleans judicial system by observing and tracking criminal court cases. Our project builds on previous capstone groups' work, focusing on data analysis, data integration, and improving Court Watch’s data collection processes. We are creating and testing models to predict bond amounts and will analyze outliers to uncover potential trends and identify missing data. We are integrating the volunteer observation data into Court Watch’s existing docket dashboard and database. Additionally, to improve data collection, we are developing an API to retrieve docket data that will be used for a data collection application Court Watch is developing.

Arie Tuckerman, Kailen Mitchell, Julia Renner

In collaboration with DAWn Audio, our team developed DAWn Audio Mobile, a cross-platform companion application designed to enhance the DAWn Audio user experience and streamline remote music collaboration. DAWn Audio’s desktop software establishes a synchronized virtual studio environment where creators can make changes together across different Digital Audio Workstations (DAWs). This solves the longstanding industry challenge of incompatible DAW software that torments the music industry. Using the frontend framework .NET Maui Blazor and the infrastructure-as-code framework SST, we built a mobile application that extends DAWn Audio’s existing architecture and impact by introducing a social networking platform, providing music creators with a dedicated forum to connect, share ideas, and coordinate projects seamlessly before transitioning to production in the desktop app. DAWn Audio Mobile serves as a centralized platform for collaboration amongst artists, removing barriers to communication and fostering dynamic, real-time creative exchange in an industry constrained by isolated workflows that hinder spontaneous creativity.

Sydney Feldman, Collette Riviere, Abby Scarry, Zachary Wiel

The New Orleans-based non-profit coalition Eye on Surveillance (EOS) works with those most harmed by government surveillance by implementing evidence-based safety systems to increase transparency between the city and the community. We have collaborated with EOS on Sawt, a Retrieval Augmented Generation-based AI chatbot created to provide the New Orleans community with easy access to answers and information discussed in City Council meetings. This year, our work includes developing an automated system that uploads all city council meetings to YouTube and adds their transcriptions to Sawt’s database for user queries to replace their current manual system. Additionally, we continue to work to add metadata to these transcripts to mitigate bias and provide context within Sawt’s answers. By streamlining processes and enhancing responses, we are contributing to EOS’s goal of helping the people of New Orleans stay informed.

Zoe Birnbaum, Daniel Cicero, Cece Haase, Gabby Reese

The City Data Group is developing an interactive dashboard that highlights reported tree hazards in New Orleans, details the severity of each report, and provides precise locations of tree damage, offering a clear and accessible view of the city's tree maintenance needs. Information is sourced from the 311 OPCD calls and Tree Location datasets from data.nola.gov, with data updated daily. Our goal is to incorporate the Tree Location data to show the nearest city-owned trees to the report. Through this feature, the user can get an idea of which tree each report most likely refers to. Additionally, our tool allows residents to upload photos and comment on specific cases, with the goals of fostering community engagement, improving data accuracy, and providing additional context. Ultimately, we hope this tool will streamline the 311 tree service response process, and give residents a better concept of where outstanding tree service requests still exist.

Rhegan Barrett, Chris Lewis

This project investigates the integration of a hybrid Concentrated Photovoltaic-Thermal (CPV-T) system for simultaneous electricity and process heat generation in Flower Hall. Key computational components include real-time data acquisition, techno-economic modeling, and automated system control. A Raspberry Pi facilitates live data transmission, solar tracking, and safety protocols. The techno-economic analysis evaluates cost savings and return on investment relative to natural gas, accounting for state-specific variables. This work aims to optimize solar cogeneration for building-integrated applications, enhancing efficiency and economic viability.

Kevin Skelly, Nicole Davis, London Jones, Ananya Anand

Operating under Dr. Brian Summa's moonshot project MAGIC-SCAN, our group investigates advanced methodologies for tumor imaging and cancer detection by integrating high-resolution microscopy with three-dimensional modeling to automate high-quality scan production. After experimenting with different scanning devices and techniques, we developed a program to automate imaging, depth mapping, and point cloud generation. Following data acquisition, we implemented meshing techniques to synthesize high-resolution 3D scans into comprehensive models. The system has been successfully tested on various inanimate objects and red meat tissue, which serves as an appropriate analog for human tissue. This work represents a significant step toward developing automated, high-fidelity imaging systems for cancer detection.

Madhangi Krishnan, Cameron McLaren

In the U.S., over 500,000 Deaf and Hard-of-Hearing (DHH) people use American Sign Language (ASL) as their primary language. However, a critical gap in healthcare accessibility persists, particularly in emergency medical services (EMS), where first responders often lack the tools and training to communicate effectively. Users will be presented various categories of terms related to emergency medical scenarios, each category having a "stack" of words and/or phrases that they will learn and subsequently be tested on. When learning each term, users will be shown a video of someone signing the word/phrase and then will be later prompted to record a video to submit to the recognition software, which is an AI-based technology capable of recognizing isolated signs. The output of this software will indicate whether or not the term was signed correctly. The website will also keep track of the user's progress using a live database so they can revisit the site in the long term. By enhancing EMS providers’ ASL proficiency, this project aims to improve emergency care accessibility for the DHH community. Future expansions may adapt the platform for other healthcare professionals, such as pharmacists, to further address communication barriers in medical settings.

Austin Nguyen, Elena Yang

The purpose of the HAYSTAC project is to detect anomalous data from simulated trajectories. Our goal is to determine abnormal movement patterns given location data over a time period of 8 weeks. We are presented a dataset of 500k agents, 100 of which are anomalous. The first part of our work focuses on developing visualizations, including timeline plots and a matrix calendar plot. The second part of our work is developing a transformer model that takes in image input and clusters data based off of movement patterns. The goal is to further be able to use these visualizations to predict anomalous behavior.

Jonathan Sears, Nick Radwin, Russel George, Zach Goodman

Musicians often encounter the challenge of replicating a desired audio tone but struggle due to the vast number of possible effect combinations and parameter settings, making brute-force methods impractical. Tone Grabber addresses this issue by employing machine learning to analyze target audio samples and predict the effects and parameters used to achieve the tone, outputted in a user-friendly interface.

Jack Lehavi, Simon Yung, Sean Hall

Our project investigates adversarial vulnerabilities in Multi-Agent Systems (MAS) and develops resilient architectures to counter agent poisoning. We introduce novel structures like the Linear Pipeline and GroupChat with self-correction loops, alongside advanced attack strategies such as Auto Bi-Injection and defenses inspired by Theory of Mind and Watchdog mechanisms. Using the highway-env framework, we simulate AI agents with memory-based decision-making in dynamic traffic scenarios, including intersections and roundabouts, while exploring the impact of poisoning retrieval-augmented generation (RAG). Our benchmarks across diverse language models ensure broad applicability in securing MAS deployments.

Gavin Galusha, Evan Nyhus, David Webster Rouqin Ji

This project seeks to develop an explainable AI-powered poker tutor that plays optimally while providing clear, human-readable reasoning for its decisions. This system will articulate the rationale behind its moves in a clear and educational manner. We train a Counterfactual Regret Minimization (CFR) agent and convert its learned strategy into a decision tree for better interpretability. To improve transparency, we incorporate Hand Rank and Draw Type features and refine our training set using PioSolver’s game abstraction research. By simplifying complex AI strategies into structured decision-making processes, we aim to make AI more accessible and understandable for users and researchers alike.

Raymond Liu, Sharon Yin, Yundan Yang, Charles Zhang

Moiré is a solar-powered sensor network and software system for monitoring abiotic conditions—soil moisture, temperature, humidity, and light—across large areas using LoRa mesh communication. Data is transmitted throughout the day and visualized via a fullstack web application.

This year, we enhanced system efficiency by implementing a hybrid Arduino/ESP-IDF firmware to reduce power consumption and improve reliability for field deployment. We also developed backend APIs, optimized database management, and refined the frontend interface for seamless data display. The system will be deployed by FCAT, an Ecuadorian NGO, to identify ideal conditions for reforestation.

Tony Tran, Brian Kisken, Sam Cohen, Drew Zimmerman

Our project, an innovative inventory management platform, will allow sellers and small businesses to make better decisions about purchasing and reselling shoes. A user will scan the barcode of a product using out system, and it will give them the live market data about that specific product from multiple websites, as well as calculate potential profits using tax data. We hope to address common problems sellers have like seeing which platform would be the best to sell their product on. Our application will integrate reselling multiple platforms to streamline the process for sellers. If they decide to purchase the product, it will also be added to their personal inventory on our platform. Our system will determine if a user can profit by buying and reselling a product.