Department of Computer Science Abstracts

Presenter(s): Elif Yaren Celebi

Showcase Advisor: Amreeta Chatterjee

Abstract: College students experience high stress and anxiety, with excessive screen time as a major contributing factor. The constant connection to devices reduces the time and space for self-reflection and emotional regulation. While many mood-tracking apps exist, they focus mainly on data collection and fail to provide insights about meaningful patterns or reduce screen time burden. These apps often require long entries and extended screen time, which can feel counterproductive and end up discouraging users from using them consistently. Objective is to promote self-awareness and help users build routines while minimizing screen exposure, we present Take5, a mood-tracking and journaling app. Through brief daily check-ins, Take5 logs user emotions, provides context-adaptive journal prompts (e.g., gratitude prompts when feeling down), and suggests simple screen-free activities like taking walks or calling friends. The app visualizes mood trends through weekly and monthly charts, helping users identify emotional patterns.

Presenter(s): Nicholas Popiel, Schuyler Denu, Onari Romain, John Evanchik

Showcase Advisor: Pradeep Atrey

Abstract: This presentation is for a capstone project in class ICSI 499. We have been assigned to create a tower defense game for AJ Picard of OneTap. This tower defense game is made in the Unity game engine and is made primarily for MacOS and Windows. The main gameplay loop will be grabbing and throwing the zombies that attack your base with your mouse, You will also be able to “cure” zombies and add them to your base for better upgrades. The game progresses in rounds that progressively get harder with more zombies. When you defeat a zombie, they drop gold you can use in the shop. At the end of each round you can build and upgrade towers and are given a “Perk” that can give a tower a special buff. You also have the ability to purchase one time upgrades and traps from the shop as well.

Presenter(s): Michael Nasierowski, Bryan Welch, Tyler Barnes

Showcase Advisor: Pradeep Atrey

Abstract: The Out-of-Home advertising industry often relies on static 2D metrics that ignore obstructions and dynamic viewing angles. To solve this, "Ad-Opt" is a 3D visualization and analytics tool built in Unity that ingests real-world GIS data to create a high-fidelity Digital Twin of Albany, NY. Users can place virtual signage within this simulation to evaluate ad space performance. 

The software utilizes a volumetric projection system, allowing billboards to "cast" a visibility footprint onto surrounding roads. This dynamically highlights road sections within the effective viewing range that remain unblocked by environmental obstacles. By sampling data points along these illuminated segments, Ad-Opt calculates an accurate visibility score. Finally, this visual data is layered with demographic and traffic flow datasets to generate a "Total Impression Score," bridging the gap between raw GIS data and actionable marketing intelligence.

Presenter(s): Srinivas Mekala, Sai Satwik Bikumandla, Mehak Seth, Dileep Reddy Chinneluka

Showcase Advisor: Jeff Offutt

Abstract: CodeCrafters is a two-player collaborative web game designed to teach foundational Python programming concepts - variables, input/output, conditionals, and functions - to beginner learners. Motivated by the need for engaging, low-barrier entry points into programming education, the game assigns players complementary roles (Architect and Builder) to collaboratively construct a virtual school campus through code. Each stage unlocks only when both players successfully complete their respective coding tasks, enforcing meaningful collaboration. The system is built using a split-screen interface pairing a live code editor with a shared visual campus, providing immediate feedback between code written and progress made. Attendees will see how game-based mechanics - role assignment, progression gates, and real-time team chat - can make collaborative programming education more engaging and accessible than traditional instructional approaches.

Presenter(s): Chaitanya Rao Alwal, Jianxiang Huang, Snehith Reddy Dropathi, Varun Reddy Karra

Showcase Advisor: Jeff Offutt

Abstract: Code Sense is an interactive web application designed to help learners strengthen their understanding of how code works by predicting program outputs. The platform presents users with short code snippets and asks them to determine the exact output produced by the program. After submitting their predicted output, the system evaluates the response and provides immediate feedback along with a clear explanation of the correct result. Users progress through a structured session that includes selecting a topic and difficulty level, answering questions, reviewing feedback, and viewing a final summary of their performance. The application is built using a React-based frontend and a FastAPI backend that handles question retrieval and answer evaluation. By focusing on code comprehension rather than code writing, Code Sense encourages learners to think critically about program behavior and helps them develop stronger debugging skills, logical reasoning, and a deeper understanding of programming concepts.

Presenter(s): Rheinard Zadanowsky, Lucas Depaola, Omolara Olaniyan, Justin Macey

Showcase Advisor: Pradeep Atrey

Abstract: Cybersecurity teams generate large volumes of raw output data from penetration tests and vulnerability scans, but transforming those results into clear, concise, and professional customer ready reports require a lot of time and manual work. In this project, our team will design, implement, and evaluate a software application that ingests penetration testing and vulnerability scan outputs, extract key information, and will convert them into a consistent structure. The application will use AI-assisted parsing/information extraction to acquire details such as severity, evidence, affected systems, and recommended fixes. Utilizing the structured data, the application will automatically generate professional customer ready reports aligned with OrbitalFire's branding. By extracting data from the penetration testing reports, refining this data to include proprietary information, and appending it to the findings report, this will drastically speed up OrbitalFire’s manual pipeline for generating reports faster, showcasing real world efficiency.

Presenter(s): Salman Khan

Showcase Advisor: Benjamin Babu

Abstract: Malnutrition in developing countries such as Pakistan, India and Bangladesh are prone to micro nutrient deficiencies such as iron, vitamin A, vitamin C, and a plethora of other essential nutrient deficiencies. This remains a critical public health challenge. However,  more food will not fix the issues, only by optimizing right kinds of food available within the strict economic constraints people there face. By demonstrating how computer science algorithms specifically variations of the knapsack problem and it’s linear programming techniques can be accurately applied to formulate mathematically optimal diets. By inputting, the exact nutritional profiles and local cost of indigenous ingredients found in the local area many nutrient dense foods. By computationally generating daily meal plans that meets all micro nutrient requirements at the minimum cost that poverty stricken people can afford. By highlighting exactly how algorithmic thinking can provide actionable data driven solution to the real world of food insecurity.

Presenter(s): Rachel Azcona, Anthony Margillo, Monzir Mekki, Jack Pfeifer, Chinwe Ofonagoro

Showcase Advisor: Pradeep Atrey

Abstract: This purpose of this project is to benchmark and test AI agents in the industry to determine which agents are best for which actions. This will help make autonomous systems flow better and improve the efficiency of industrial systems. To do this, we utilize a Model Context Protocol (MCP) server to make IBM’s existing AssetOpsBench more flexible while maintaining its efficient benchmarking process.

Presenter(s): Jyotsana Parkhedkar, Harshini Narra, Rishikesh Sirisilla, Prathamesh Kale

Showcase Advisor: Jeff Offutt

Abstract: AtlasDash is an interactive web-based educational geography game designed to make learning world geography engaging and accessible for users of all ages. The application challenges players through drag-and-drop map puzzles, flag identification exercises, and region-based challenges that reinforce geographical awareness in an intuitive, game-like environment.

AtlasDash delivers a responsive, multi-device experience with a real-time scoring system that tracks accuracy, attempts, mistakes, and completion time. Players can filter challenges by continent, region, or difficulty level, while an informational panel provides cultural and geographical context for each country or region explored.

With a clean, user-friendly interface, AtlasDash bridges the gap between education and entertainment, transforming passive geographical learning into an active and rewarding experience. Deployed as a publicly accessible web application, the platform invites users worldwide to strengthen their understanding of global geography in a fun and visually immersive way.

Presenter(s): Aharon Leverton, Sophia Giler, Eli Pardo, Renald Mendez, Keziah Job

Showcase Advisor: Pradeep Atrey

Abstract: Our Chauffeur App provides a seamless and reliable way to request professional transportation whenever you need it. Designed for convenience and efficiency, the app connects riders with experienced chauffeurs who deliver safe, comfortable, and punctual rides. Users can easily schedule rides in advance or request one on demand, track their driver in real time, and receive accurate arrival updates directly through the app.

The platform focuses on premium service, ensuring that every ride meets high standards of professionalism and customer care. With secure payments, clear pricing, and an intuitive interface, booking transportation becomes quick and stress-free. Riders can also view driver details, trip history, and ride confirmations all in one place.

Whether you need transportation to the airport, a business meeting, a special event, or simply around the city, the Chauffeur App offers a dependable solution that prioritizes comfort, safety, and reliability for every journey.

Presenter(s): V R KKautilya Shiva Kumar Miryala, Venkata Manikanta Prem sai Potukuchi, Kumara Rama Maruthi Kandikonda, Vignesh Ponnam

Showcase Advisor: Jeff Offutt

Abstract: Code.py is an interactive educational platform designed to improve programming comprehension by training users to mentally execute code and predict program outputs before execution. The system focuses on two widely used programming languages, Python and JavaScript, and provides curated exercises categorized by language, difficulty level, and programming topics. Users select an exercise, analyze the code snippet, and submit their predicted output along with a confidence level indicating their certainty. The platform then executes the code in a controlled environment and compares the predicted output with the actual result using a similarity scoring mechanism. Based on this evaluation, the system provides immediate feedback, including correctness indicators, output comparison, and explanatory guidance to reinforce conceptual understanding. Additionally, progress dashboard tracks performance metrics such as total attempts, accuracy, and learning streaks to help users monitor improvement over time. By emphasizing prediction-based learning and instant feedback, Code.py promotes programming intuition and strengthens users’ ability.

Presenter(s): Avi Chauhan, Jainik Desai, Ishan Pathak, Rahil Shah, Karan Patel

Showcase Advisor: Jeff Offutt

Abstract: CoderV: A Visual and Interactive Platform for Intuitive Coding Education 

CoderV is an innovative coding education platform designed to transform the way learners particularly beginners and young students engage with programming concepts. Unlike conventional text-heavy learning tools, CoderV leverages a node based visual interface that illustrates how code components interconnect and function as a unified system, making abstract logic tangible and intuitive. The platform introduces gamified, interactive learning mechanics such as fill in the blank exercises, drag and drop indentation challenges, and visual code flow mapping to reduce the intimidation barrier traditionally associated with programming. By presenting code as a living, visual network rather than static syntax, CoderV fosters deeper conceptual understanding and sustained engagement. CoderV’s core mission is to reframe coding not as an academic obligation, but as a creative and enjoyable activity making computational thinking accessible, visual and fun for learners at every stage of their journey.

Presenter(s): Florian Charles, DeAndre Lawson, Jaden Baker, Conner Chew

Showcase Advisor: Pradeep Atrey

Abstract: This project extends CloudLens, a mobile application originally developed by a Spring 2025 capstone team, through the design, implementation, and evaluation of a credibility-assessment framework for multimodal social media content. The original system was designed to capture images using a mobile device, transmit them to a cloud-based service for processing, and return the results to the user interface. Building on this foundation, the present work expands the application’s functionality by enabling the analysis of screenshots and similar images containing both textual and visual elements. The proposed enhancement focuses on developing an algorithm capable of assessing the truthfulness of such content by examining informational cues in text and imagery. By integrating this functionality into the cloud-based architecture, the enhanced CloudLens application aims to provide users with a tool for evaluating the credibility of media posts and addressing digital misinformation through automated verification technologies.

Presenter(s): Medha Gajula

Showcase Advisor: Sherry Sahebi

Abstract: As AI systems become central to how we find and use information, it is important to understand how they influence human perception and decision-making. Human-Computer Interaction (HCI) studies how people communicate with digital systems, aiming to make technology more efficient, safe, and user-friendly. A common example today is Conversational AI such as ChatGPT and Claude AI. These systems are built on neural network models that identify patterns in data to generate responses. This study examines whether conversational AI influences how users perceive the quality and usefulness of information compared to traditional search engines. A small user study with approximately 20 college students will involve completing information-seeking tasks using both a conversational AI system and a traditional search engine. Participants will then evaluate each system based on quality, usefulness, preference, and trust. These findings may help researchers and designers improve HCI and inform future design of AI-driven interfaces.

Presenter(s): Dylan Tarace, Evan Wojtalewski, Noah Clough, Corey Jacobowitz

Showcase Advisor: Pradeep Atrey

Abstract: Functional Dependency Network Analysis (FDNA) models infrastructure systems as networks in which each node’s operability depends on the condition of neighboring nodes. Node health is represented by a value between 0 and 1 and calculated using a weakest-link rule weighted by parameters α and β. Estimating these parameters accurately is difficult when infrastructure data is distributed or sensitive.

This project develops a Python simulation that applies federated learning (FL) to improve FDNA operability prediction without centralizing data. A distributed dependency network is generated and node operability evolves over time through simulated failures. An LSTM neural network learns patterns in recent operability history and predicts α and β values. Multiple clients train the model locally and share model updates with a central server, which aggregates them using the FedAvg algorithm.

Performance is evaluated using MSE, MAE, R², Accuracy, F1 score, and AUROC, and compared with centralized, isolated, and fixed-parameter baselines.

Presenter(s): Vincent Jiang, Kevin Chen, George Lee

Showcase Advisor: Pradeep Atrey

Abstract: This project proposes a web application that helps small and medium-sized businesses manage and monitor their financial activities. The system will provide secure user authentication, tools for manual data entry, and the ability to upload receipts or invoices. Financial data will be stored and organized to allow easy review and management. A key feature of the application is an alert system that identifies unusual spending patterns based on predefined rules. The home page will include a dashboard summarizing overall spending trends and displaying any unresolved alerts. Users will be able to add financial data through file uploads or manual input, after which the information can be reviewed and corrected. The system will then categorize transactions automatically using default rules or through user-defined classifications, helping businesses maintain organized and transparent financial records.

Presenter(s): Irien Akter

Showcase Advisor: Pradeep Atrey

Abstract: Social media platforms face escalating harm from concentrated waves of toxic interactions, which we term Neg Storms. Unlike isolated abusive remarks, these storms emerge through rapid, correlated actions that amplify negativity and create severe risks for targets and communities. Existing moderation approaches largely focus on detection of individual comments, missing the situational dynamics that drive escalation over time. This paper introduces a proactive framework for forecasting neg storms using early conversational signals. We formalize Comment Storm Severity (CSS) as a time-aware metric of thread-level toxicity, propose models that predict CSS from only the first k comments, and evaluate feature sets combining timing and content cues. Experiments on Reddit and Instagram show that timing features alone outperform content-only features, and that combining both yields the best performance (ROC-AUC ≈ 79.7%; R² ≈ 0.24). While predictive scores are modest, results validate the harmful situations before they fully unfold and support proactive moderation.

Presenter(s): Alden Strafford, Jay Patel, Narayani Khatavkar, Benjamin Babu

Showcase Advisor: Jeff Offutt

Abstract: 2D 2 player local co-op sprite based adventure game for grades K-5 that has users complete minigames related to math, science, and reading in a grove based adventure map in order to find the 3 ancient gems and restore life to the Great Tree.

Presenter(s): Jawad Hossain

Showcase Advisor: Chong Liu

Abstract: Small language models often struggle with complex mathematical reasoning due to limited capacity to maintain long chains of intermediate steps and to recover from early errors. We address this challenge by introducing a hint-assisted reasoning framework that incrementally guides small language models through multi-step mathematical problem solving. The proposed approach decomposes solutions into sequential reasoning steps and provides context-aware hints, with each hint generated conditionally on the problem statement and the accumulated reasoning history. This stepwise, conditional guidance reduces error propagation and enables models to focus on localized subproblems without revealing full solutions. Experiments across diverse mathematical benchmarks demonstrate that hint assistance consistently improves reasoning accuracy for small language models, yielding substantial gains over standard prompting while preserving model efficiency. These results suggest that structured and local hinting is an effective and lightweight mechanism for boosting mathematical reasoning in small language models.

Presenter(s): Bhavi Patel, Blesson Binoy, Jenny Chen, Jackie Wu, Tahsina Mahdiah

Showcase Advisor: Shashank Arora

Abstract: Finding jobs and internships often requires searching across multiple platforms such as LinkedIn, Indeed, and Handshake, which can make the process fragmented and difficult to manage. HireMe is a centralized job discovery and application tracking platform designed primarily for students and new graduates. The system aggregates job listings from multiple sources and presents them through an interactive map-based interface, allowing users to visualize opportunities geographically and filter them based on location, role, and preferences. In addition to job discovery, HireMe provides an application tracking dashboard where users can organize saved positions, monitor application progress, and manage deadlines. By combining job aggregation, geographic visualization, and application tracking in one platform, HireMe aims to simplify the job search process and help users stay organized while navigating the competitive job market.

Presenter(s): Jeremy Varghese

Showcase Advisor: Aishwari Talhan

Abstract: Mobility impairments are widespread across orthopedic, neurological, and deconditioned patient populations, often requiring long periods of structured rehabilitation. Traditional lower limb rehabilitation can be repetitive and insufficiently engaging, which reduces patient motivation and limits functional progress. This project presents an Extended Reality rehabilitation system integrated with the Vector Gait and Safety System to deliver immersive, gamified therapy supported by clinical grade tracking. The system features hands free interaction, real time performance monitoring, and automatic data logging. An in house large language model evaluates early performance metrics and adjusts task difficulty to create a personalized and adaptive therapy experience. Four custom mixed reality tasks target reach and reaction, hand eye coordination, dynamic balance, and weight shifting. Testing with stroke, traumatic brain injury, and rehabilitation patients demonstrated high engagement, measurable improvements in gait speed, and strong interest in personalized content. Future work includes developing a virtual AI assistant and standardized assessment integrations.

Presenter(s): Xianqi Deng

Showcase Advisor: Petko Bogdanov

Abstract: Masked language modeling (MLM) is widely used for molecular sequence pretraining, but standard uniform masking underutilizes training budget and underexposes rare yet chemically meaningful patterns. We propose a difficulty-aware masking framework that replaces uniform masking with a learnable policy trained to generate challenging masked examples. Our method introduces a teacher-guided generate-and-verify workflow that jointly trains a masked language model and a contextual bandit masking policy. An EMA teacher produces token-level hardness signals that condition a policy network to propose candidate mask subsets using token identity, positional features, and difficulty cues. The teacher evaluates each candidate via reconstruction loss and selects the hardest mask for student training, while the policy is updated with REINFORCE to favor high-reward subsets. This set-level formulation captures dependencies among masked tokens and emphasizes long-tail chemical patterns. Experiments show improved robustness under rare-token distribution shifts while maintaining competitive in-distribution performance.

Presenter(s): Blessing Okoro

Showcase Advisor: Mariya Zheleva

Abstract: Traditional spectrum analytics focuses either on transmitter time-frequency characterization or localization. Solving these tasks in isolation has limited applicability to complex real-world scenarios where both time-frequency and spatial awareness are critical for sharing and enforcement. In addition, the potential accuracy gains from joint characterization and localization have not been explored. To address these gaps, we propose MDL (Multi-vantage Detection and Localization), a framework for joint transmitter characterization across time, frequency, and space using multi-vantage point traces. We model multi-sensor spectrum measurements as a three-way tensor, where slices represent Power Spectral Density (PSD) data from individual sensors. We employ sparse dictionary-coding with appropriate analytical dictionaries for the temporal, spatial and frequency modes to decompose the input tensor and utilize the extracted factors for characterization and localization. Our method simultaneously estimates time-frequency occupancy patterns and aproxy Received Signal Strength (RSS) for multiple transmitters, enabling joint detection, separation and localization.

Presenter(s): Krishna Sree Guguloth, Shreyash Govind Mungilwar, Sahith Kumar Reddy Mogusala, Abhishek Santhakumar, Nikhil Shetty Lakshmishetty

Showcase Advisor: Jeff Offutt

Abstract: Knowledge Things is a cooperative multiplayer quiz game for children in Grades 3 to 5, ages 8 to 11. It covers six subjects: Mathematics, Science, English, History, Geography, and General Knowledge. Topics and difficulty are adjusted to the chosen grade. Two to four players join using a room code, enter a display name, and answer questions together. Validation happens only after everyone has submitted. Questions are generated by AI and get harder or easier based on how the team performs. The system provides hints and explanations when answers are wrong. Players see a shared team score, an individual leaderboard with speed bonuses, and a timer for each question. The platform is built with Next.js, Express, Socket.io, and TypeScript, and can be deployed to the cloud. Knowledge Things demonstrates how cooperative, AI-assisted games can support interactive learning and peer collaboration in primary education.

Presenter(s): Minseon Kim

Showcase Advisor: Pradeep Atrey

Abstract: This research project examines how Korean AI companies are preparing for emerging global AI governance and regulatory frameworks, particularly the EU AI Act and the NIST AI Risk Management Framework. The study focuses on evaluating the technical and organizational readiness of Korean AI startups that are developing large language models, generative AI applications, or AI infrastructure. Through structured interviews with technical leaders and engineers, the project investigates current practices in AI safety testing, risk management, data governance, and compliance preparation. Each participating company receives a brief technical assessment highlighting potential compliance gaps and recommendations for improvement. By aggregating these findings, the research aims to map the current state of AI governance readiness in the Korean AI ecosystem and identify common challenges faced by companies seeking to expand into international markets.

Presenter(s): Ahmed Aslam, Farhad Safi, Akhil Paulson, Muwahid Seraj

Showcase Advisor: Pradeep Atrey

Abstract: This project involves the continuation and new development of an ongoing application that manages and optimizes energy usage for various building types, such as houses, hospitals, and schools. The backend features an SQL server for data storage and retrieval, Java SpringBoot as the framework, and algorithms to evaluate and improve energy consumption. The frontend uses React to provide a user-friendly interface and is hosted on Amazon Web Services (AWS) to ensure scalability and public accessibility. This project consists of adding new building investment models to the backend and front end. Fixing several issues left over from the previous project team and developing several AI-based modules to improve testing, performance, and increase capability. This project includes work with a startup entrepreneurial company to help increase clean energy development. This project is meant to make it easy for users to understand green upgrades on their buildings.

Presenter(s): Ahmed Al-Mashraie

Showcase Advisor: Amreeta Chatterjee

Abstract: Older adults managing multiple conditions frequently experience preventable adverse drug interactions due to tiny print and confusing medical jargon on over-the-counter labels. MediTrust is a mobile application that acts as a "Digital Pharmacist". Moving away from a text-heavy "search engine" feel, MediTrust utilizes a simple chat interface. Users can scan a medication bottle or ask a question via voice or text. The application then cross-references the drug against the user's persistent "Health Profile" (including active prescriptions and allergies) using established medical databases like the FDA API. To eliminate cognitive overload, the app delivers a clear, binary safety verification. By prioritizing multimodal interactions as large icons and voice-guided input, MediTrust bridges the critical "After-Hours" gap when medical professionals are unavailable.

Presenter(s): Ariana Nieves, Allison Talla, Evan Budhoo, Derya Ertik, Malka Syed  

Showcase Advisor: Pradeep Atrey

Abstract: This project focuses on modernizing a legacy web application by migrating its backend from Node.js 4, an end-of-life runtime, to Node.js 24, a current long-term supported environment. The upgrade aims to improve security, performance, maintainability, and compatibility with modern development tools and libraries. Because the existing system relies on outdated dependencies and potentially deprecated APIs, the migration will follow a staged approach that incrementally upgrades Node.js versions while identifying and resolving compatibility issues. Our team will analyze the current codebase, update or replace unsupported packages, refactor code where necessary, and ensure that native modules and build tools remain functional throughout the transition. The outcome will be a stable backend running on Node.js 24 that preserves the functionality of the original system while enabling future development and long-term support.

Presenter(s): Alessandro Marino Calzado El Gornati

Showcase Advisor: Ming-Ching Chang

Abstract: Epileptic seizure detection traditionally relies on manual inspection of electroencephalography (EEG) signals and behavioral recordings, which is labor-intensive and difficult to scale for large experimental datasets. In collaboration with Albany Medical Center, this project constructs a synchronized multimodal dataset combining long-duration rodent behavioral video recordings with simultaneously recorded EEG signals and expert seizure annotations.

The dataset consists of multi-terabyte video recordings paired with high-frequency EEG signals and timestamp logs. To enable computational analysis, we developed a preprocessing pipeline that converts proprietary binary timestamp files into structured formats, segments large recordings into manageable chunks, and aligns these segments with corresponding signals. Handwritten seizure annotations are digitized and synchronized with the data streams to produce ground-truth labels.

This dataset forms the foundation for automated seizure detection research. Baseline machine-learning models will classify seizure events using combined behavioral and EEG features, with future work exploring multimodal deep learning approaches for scalable neurological analysis.

Presenter(s): Zakaria Shams Siam

Showcase Advisor: Chong Liu

Abstract: In many scientific screening problems, the goal is not to find just one "best" candidate, but to identify a small set of promising candidates that meet several requirements while representing a broad range of desirable outcomes. This is especially important in areas such as drug discovery and materials design, where downstream validation is expensive and decision-makers need diverse, feasible options. In this work, we define a new problem, called multi-objective coverage, that asks how to efficiently find representative candidates that meet multiple requirements while spanning a broad range of acceptable outcomes. To address this new problem, we present a new method, MOC-CAS, that selects candidate designs to uncover a broad range of acceptable outcome patterns. This focus differs from standard multi-objective methods that mainly chase the Pareto front. On large protein-target datasets related to cancer and SARS-CoV-2, the method finds stronger feasible coverage than competitive baselines under similar evaluation budgets.

Presenter(s): saba irfan, Stacey Gao, Maurice D Ogada, Sahim M Hussaini, Kameron C Robinson

Showcase Advisor: Pradeep Atrey

Abstract: Neuro Connect is a micro-frontend-based healthcare platform that integrates multiple independent health tracking modules into a unified host dashboard. This milestone presents the proposed solution architecture, covering micro-frontend integration using Native Federation, client-side authentication and authorization, module interaction and data flow, and design tradeoff analysis. The platform is implemented using Angular for client-side rendering, Native Federation for runtime module loading, and simulated health data with no backend server.

Presenter(s): Ishrat Jahan Mohima, Clark Mattoon

Showcase Advisor: Mariya Zheleva

Abstract: Reliable, high-quality data is required for machine learning models and efficient spectrum sharing. However, internal temperature fluctuation and undetected gain drift can alter data in Software Defined Radios (SDR). This creates a "blind spot" where analysts fail to ensure trustworthiness of data without expensive lab equipment.  
Employing Broadcast FM as a trusted, high power Signal of Opportunity (SoOP), we provide a framework for "on the fly" sensor health tracking. Evaluating real-time hardware performance with conventional FM signal characteristics, our approach executes real-time sensor calibration. We ensure data fidelity through three key metrics which are veracity, intermittency and ambiguity.

To prevent data loss and inaccuracy, "on the fly" sensor health can be monitored and sensor setups can be dynamically optimized using our suggested framework. Along with ensuring sensing integrity and trust in automated wireless systems through the incorporation of radiometric concepts, our method is also lightweight and reproducible.

Presenter(s): Dominic Fontana, Jack Schwabe, Lakshay Bansal, Joe Balzo, Matthew Rowe

Showcase Advisor: Pradeep Atrey

Abstract: This project involved designing and implementing a full-stack application that verifies the authenticity of digital images by analyzing the consistency between visual content and embedded camera metadata (EXIF). The application is usable as a web and mobile solution that allows you to input files to then be determined if their metadata is either accurate or suspicious. Past results are saved in a database for users to look at after running the algorithm that detects the falsified images. The previous uploads are stored behind a user authentication system.

Presenter(s): Simge Sahiner

Showcase Advisor: Sherry Sahebi

Abstract: Understanding when and how students procrastinate helps improve learning and design better tools. In this project, I use data from the Proccoli app to study student procrastination using activity logs. The dataset includes records such as when students open assignments, how often they study, when they complete tasks, and goals they set for themselves.

The project includes cleaning and preparing data, exploring the dataset, creating features, and applying supervised and unsupervised models. I focus on time-related features like gaps between actual study time and the time students set as a goal, and study intervals.

Clustering methods find groups of students with similar study patterns, while supervised models predict whether a student might procrastinate in the future based on past activity.

I also create visualizations to show patterns in engagement and how procrastination relates to performance. The goal is to better understand procrastination based on the subject, and assignment type.

Presenter(s): Vaasu Taneja

Showcase Advisor: Mariya Zheleva

Abstract: Efficient spectrum sharing hinges on mutual awareness of coexisting technologies, which in turn, requires reliable analytics. At the heart of the process is RF spectrum measurements – IQ timeseries or power spectral density waterfalls, which are inherently noisy. To aid analytics, RF denoising has been considered as a pre-processing step. This paper presents a comparative analysis of three denoising methods: classic wavelet-based denoising, deep learning convolutional autoencoders, and a new stream stemming from denoising diffusion models. We explore the effects of denoising on two awareness applications: active/idle band detection using thresholding, and a recent machine learning based transmitter separation and characterization method that uses sparse dictionary coding. We show that all denoising can offer substantial advantages with simple thresholding. However, the benefits are not as clear when denoising is mixed with machine learning based analytics.

Presenter(s): Zakaria Shams Siam

Showcase Advisor: Chong Liu

Abstract: Many important discovery and design problems require finding the best option when each evaluation is expensive, noisy, and treated as a black box. Examples include drug discovery, materials design, and automated model tuning, where even a small reduction in the number of trials can save substantial time and cost. This work studies how quantum computing can help accelerate that search, especially in high-dimensional settings where existing methods often become impractical. We present a new algorithm, Q-NLB-UCB, that combines quantum estimation with parametric function approximation to guide sequential decision-making more efficiently. The method is designed to remain effective even when the input space is very large, making it more suitable for realistic scientific applications. Experiments on synthetic benchmarks and real-world hyperparameter tuning tasks show that the approach can outperform existing state-of-the-art quantum baselines in both search efficiency and runtime.

Presenter(s): David Aduku, Leo Meurs, Bryan Rosales Elias, Justin De Paula

Showcase Advisor: Pradeep Atrey

Abstract: The SecureTrust Escrow App is a proposed software platform designed to increase trust and security in online peer-to-peer transactions. Many buyers and sellers hesitate to conduct transactions for goods or services due to concerns about fraud, non-delivery, or payment disputes. This project addresses those challenges by implementing an escrow-based system where funds are securely held by a neutral third party until agreed-upon conditions are satisfied. The application will support secure fund deposits, user verification, milestone-based transaction workflows, and conditional fund release to ensure accountability for both parties. The system will be designed using modern web development technologies, secure database management, and cybersecurity best practices to protect sensitive financial data. This first phase focuses on designing and implementing the core escrow architecture and user experience, while evaluating legal, ethical, and usability considerations for building trustworthy digital marketplaces.

Presenter(s): SHARMILEE RAJKUMAR RAJAN

Showcase Advisor: Pradeep Atrey

Abstract: The rapid rise of AI-generated and edited images has made it increasingly difficult to distinguish authentic photographs from manipulated content. Digital image forensics addresses this challenge by detecting inconsistencies between image 
content and metadata. Among various forensic cues, the exposure 
triangle parameters namely ISO Speed Ratings (ISO), aperture (F number), and shutter speed offer a physically grounded reference for verifying authenticity. We propose a physics-guided latent triad regression framework that predicts these parameters directly from image pixel content while enforcing exposure value (EV) 
consistency through the exposure equation. Our model predicts ISO, F-number, and EV in log space, deriving shutter speed to ensure physically coherent and non-redundant predictions. By embedding physical exposure laws into learning, the framework produces interpretable, exposure consistent predictions that enhance metadata verification, camera provenance analysis, and image authenticity assessment.

Presenter(s): Brandon Schadoff, Arsceem Newkirk, Beatrice Dugan, Easha Mashud

Showcase Advisor: Pradeep Atrey

Abstract: Developing a modular Smart Home Terminal, this compact Raspberry Pi device features a touch-optimized UI inspired by the Nothing/CMF aesthetic. Utilizing a 1x1 and 2x2 tiled widget system, it offers a high-performance, low-cost alternative to proprietary ecosystems by leveraging open-source hardware.

The architecture relies on a Python-based Flask backend to serve a local configuration page for real-time adjustments. Data persistence is handled via SQLite, offering a self-hosted option that secures preferences and plugins without external server privacy concerns.

To ensure broad compatibility, the device implements the Matter protocol and a custom synchronization plugin. Furthermore, the system can leverage AI to analyze environmental data from sensors and execute proactive automations, delivering a responsive smart home experience. The goal is to provide a self-hosted, at-a-glance data experience without distractions.

Presenter(s): HaoJian Huang, Chaoji Yang, JunWei Zhuo, Dylan Walker, Athulya Mathew

Showcase Advisor: Pradeep Atrey

Abstract: Talk Assist is a Flutter-based mobile application providing a voice-to-voice interface for visually impaired users. This iteration introduces a robust offline-first approach, ensuring accessibility regardless of internet connectivity. The system utilizes a hybrid STT to LLM to TTS pipeline: while it defaults to high-capacity online models via API call to providers for enhanced performance, it seamlessly transitions to a local model when offline.

By leveraging sherpa_onnx to run Whisper Tiny and Piper locally, the app maintains core Speech-to-Text and Text-to-Speech functionality on-device. This architecture prioritizes user privacy and reliability. Currently optimized for Android, Talk Assist focuses on accent handling and environmental adaptation, evaluated through real-world scenarios to refine its effectiveness as a versatile, dependable assistive tool for visually impaired community.

Presenter(s): Elham Sadeghi, Jason Scheffel, Garima Yadav, Xianqi Deng

Showcase Advisor: Petko Bogdanov

Abstract: Predictive modeling for natural science sequences—such as peptides, DNA nanoclusters, and molecules—often relies on Pretrained Foundation Models (PFMs). Although PFMs produce rich contextualized token embeddings, a key “readout” bottleneck remains: common aggregation methods (e.g., mean or max pooling or CLS tokens) compress embeddings into a single vector and can discard non-local relational information. We propose Token Embedding Geometry (TEG), a framework that exploits the latent geometry of token embeddings for improved sequence-level prediction. TEG constructs geometric graphs from embedding-space proximity and integrates them with sequence order. Graph Neural Network layers and graph-based pooling methods (such as MinCut and GMT) preserve structural dependencies that simple pooling ignores. Experiments across chemical, biological, and synthetic sequences show that TEG outperforms standard baselines and captures non-local motifs in DNA-stabilized silver nanocluster fluorescence that are often lost in global averaging, highlighting the predictive value of latent embedding geometry without requiring external 3D coordinates.

Presenter(s): Emily Homrighaus, Zhisong Chen, John Young, Sky Van Laan

Showcase Advisor: Pradeep Atrey

Abstract: Tran is a new compiled object oriented programming language with the potential to rival the likes of Java and C#. Currently Tran is a fully functioning language but it still lacks some basic quality of life features. Among these missing features is a debugger. TranSquash aims to be a full fledged visual debugger for Tran that will integrate seamlessly into the Tran IDE. TranSquash features include breakpoints, stepping, and variable watches. One feature TranSquash includes that many debuggers for similar languages lack is the ability to see a variable’s previous values and when the variable changed to that value. TranSquash also has an easy to use design that still offers the power and features required for a modern debugger.

Presenter(s): Cyril Asante, Udesh Goberdhan, John Vasta, Zachary Roundy

Showcase Advisor: Pradeep Atrey

Abstract: The Small Business Visibility Audit Engine is a web-based tool designed to help small businesses understand how easily customers can find them online. Online visibility includes factors such as search engine presence, Google Business Profile setup, social media activity, professional email configuration, and website speed and usability.The system will automatically review key elements of a business’s digital presence and provide clear analytics along with practical recommendations for improvement. It will generate a standardized Visibility Score that allows businesses to quickly understand their current online performance and track progress over time.Inspired by tools like Google Lighthouse and SEMrush, the platform evaluates website quality, search visibility, and overall digital readiness using automated checks and structured scoring. The goal is to deliver reliable, easy-to-understand insights that help small businesses improve discoverability, attract more customers, and strengthen their online presence.

Presenter(s): Vamsi Muppana, Sindhu Sameera Gogulapati, Ravi Shankar Bhange, Dedeepya Dora, Varun Krishna Kajjam

Showcase Advisor: Jeff Offutt

Abstract: This project presents a web-based educational tool designed to help students improve their ability to read and understand Python code. The system provides predefined Python code snippets and challenges users to analyze the code and predict its output. Users can manually execute code, submit answers, and receive immediate feedback on their responses. The platform supports multiple difficulty levels to accommodate learners with different levels of programming experience. It also tracks user progress during each session to help learners monitor their improvement. The application is built using a modern web architecture with React for the frontend, FastAPI for the backend, PostgreSQL for data storage, and Python as the core programming language. The system is deployed on AWS to ensure accessibility through a web browser. This tool aims to enhance programming comprehension skills through interactive and structured practice.