CSE 2025 Fall Departmental Demo Day
We're thrilled
to invite you to the thirteenth bi-annual Comp. Sci. & Eng. Fall Demo Day. Student groups from several CSE capstone classes will be presenting the culmination of 3-months of effort, hard work, (metaphorical) blood, sweat (well... caffeine really), and tears (see above).
Where: Davis Hall; 1st and 2nd Floor Atrium
When: Monday, December 8th, 2025
Schedule
2:00 PM -
Event opens
2:05 PM -
Opening remarks
2:10 PM -
Judging starts
4:00 PM -
Judges will get together to review results and pick winners
4:30 PM -
Winners announced and videos played for the winning teams
5:00 PM -
Event closes
Setup
Setup will start at Noon. Tables will be available for both sponsors (each sponsor will get a table) and demo participants (2 to a table). Easels will be available for participants. If you need power, please let us know! If you have any other special requests, please contact ahunt@buffalo.edu to let me know, and we will do our best to accomodate you. There are two hours reserved for setup - you can come at any time during that period to get organized, but please make sure you leave yourself enough time to be ready to go by 2PM, to give you the chance to network.
Networking
Before we open the atrium to students and the public, we’ll have some time reserved for the participants to come and chat with the sponsors and the judges. Pizza will be there as well (A big thank you to our sponsors!), so that the participants and sponsors can have a chance to eat before demos begin!
Judging
During the demo, judges will circulate to the participants demo stations, and they will be rating each project on a specific set of criteria. Judges, expect to spend approximately five minutes with each team, in order to give you time to see them all. You will be assigned a set of projects to view specifically, but you can feel free to talk to more teams as time permits! Teams, keep this in mind and keep your presentations crisp and to the point!
Prizes
There will be prizes for the top teams selected by the judges. They will be announced in the atrium, and there will be a quick photo op for each winner. Good luck to everyone, and I can’t wait to see you all there!
Winners
1st Place - TBD
2nd Place - TBD
3rd Place - TBD
Acknowledgements
Sponsors
Judges
- Chris Miller
- Nicholas MacRae
- Sai Roshan Ayyalasomayajula
- Joe Forsyth
- Hannah Wilcox
- Jesse Hartloff
- Tyler Poland
- Mary Ruelle
- John Ebel
- Luke Gardner
- Karlene Kardysauskas
- Brianna Swartz
- Jeremy Jones
- Nick Branholm
- Pratik Pokharel
- Nick Minor
- Jim Brandt
- Nitin Kulkarni
- Nemo Dighe
- Mike Canzoneri
- Naresh Devulapally
- Xiangyu Guo
- Matthew Hertz
- Andrew Lavoie
Presented Projects
( CSE587: Data Intensive Computing)
- Amazon Books Review Analysis Using PySpark & Machine Learning
-
This project builds a complete end-to-end big-data analytics pipeline to process and analyze the multi-million record Amazon Books Reviews dataset using Pandas, PySpark, and machine learning. The workflow includes extensive data cleaning, distributed preprocessing, and exploratory analysis of review patterns such as rating distributions, review lengths, and reviewer behavior. Four ML tasks were implemented at scale, sentiment classification with Linear SVM, helpful-vote prediction using Poisson regression, reviewer segmentation via K-Means, and extractive multi-review summarization using Random Forest regression with transformer-based features, producing robust insights and scalable models for understanding book reviews at large scale.
(CSE 302/303/402)
- Unfold Studio
-
Unfold Studio is a free, web-based learning platform created to support students who find coding intimidating or lack access to programs that teach it, as many existing platforms are too technical or not designed for beginners and social science learners. This creates a barrier for students who could benefit from coding as a way to express creativity and build programming skills. Educators also lack tools to understand or observe how students are using AI, and many students are copy-pasting from AI platforms instead of learning how to think through problems or write on their own. Unfold Studio provides a low pressure, creative environment where students can experiment and gradually learn coding concepts while developing logic and problem-solving skills. The platform combines storytelling, coding, and AI, and students stay in control of when and how AI supports them. It is designed for students and educators in multiple fields, including social sciences, CS, and middle and high school learners, with a primary focus on introductory CS courses. Its flexibility also allows educators in English, Social Studies, and CS to use it for interdisciplinary learning across different fields.
- LED Video Wall
-
Our team continued development on an LED Video Wall originally built by a previous group, adding several new interactive features. We created a web-based interface that allows users to easily control the system, including uploading images that display directly on the physical LED panels. The website also includes controls to start and stop the LED Video Wall remotely. These additions highlight a seamless connection between the web interface and the microcontrollers powering the LED system, creating an intuitive and engaging user experience.
- UB Hacking CMS
-
A comprehensive hackathon management platform built for UB Hacking, designed to replace manual processes with automated digital solutions. This platform enables organizers to efficiently manage registrations, check-ins, communications, and logistics without requiring technical expertise.
- Customer Atlas: A Map-Driven CRM for Outreach
-
This project focuses on building an interactive visualization tool that displays a geographic map highlighting customers in specific areas who are interested in products listed on the website. The system includes a variety of dynamic filters that allow users to pinpoint areas of interest, compare regions, and track engagement across multiple products simultaneously. An adaptive onboarding-friendly database helps new users easily add and manage product information, while integrated interest-tracking features automatically record customer interactions and preferences. The tool also supports live-time updates pulled directly from internal documentation, ensuring the visualization always reflects the most current customer activity and product data.
- Pirouette
-
The first high-order functional language for writing Distributed Systems
- AI Energy Optimization(GreenAI)
-
The GreenAI team profiles ML workloads at a method level, identifies energy hotspots, uses ML to generate optimization suggestions, and automatically refactors code to reduce CPU/GPU energy usage while keeping performance the same.
- Encryted storage
-
This project provides a secure data-storage system where a user’s encryption key is split into secret shares across multiple devices, preventing any single compromised device from exposing protected data. Devices authenticate each other using their own public-private key pairs and establish pairwise encrypted channels through authenticated Diffie-Hellman. A server (via NTFY) is used only as a relay for encrypted messages, ensuring that all sensitive content remains end-to-end protected between trusted devices.
(CSE 420 : Game Design)
- Goose Mayhem
-
Go wild! Play as a Canadian goose whose only goal is to cause mass destruction within a city. Unlock powerful abilities that cause more destruction and avoid any enemies that will try to stop you.
- Demon Depths
-
A souls-like video game with demons, metal, and lots of "heck yeah" vibes.
- Nona
-
A cozy, 2D platformer where you play as a cat trying to return a precious item to their owner. Traverse through a big city and reach your owners office to save the day!
- TaserTaxi
-
You are tasked with capturing escaped animals that roam around the city. However, you are given a limited time to do so and if you don't drive according to traffic laws, your car will explode on the spot and you fail.
- Sugar World 2.0
-
A 3D third-person fighting Game, players control characters fighting with AI enemy by attacking and abilities
- Lost Of Castle
-
This is a new take on tower defense, now in full 3D. You play as a commander who can both control the battlefield and jump into the fight yourself. With modern weapons, you can run around, help defend, and change the battle when it matters most. This mix of planning and fighting makes the game exciting and immersive.
- Escape from UB
-
basically 5 nights at Freddys but in UB, Steal the mid term paper answers while an anamatronic is trying to stop and kill you. Use cameras doors and flashlight to buy you enough time to win.
- Cat Fish Island
-
A game about fishing and cats! Inspired by Stardew Valley and Animal Crossing, this game set on an island in the sea, secluded from civilization, with only a few cats and cat people living there. It's a game meant to appeal to the more casual side of modern games. Through fishing and rearing cats, the player progresses and obtains currency and upgrades. During this journey, they follow the story, gain cosmetics, and unearth the mysteries of Catfish Island, step by step.
- Anko's Loop Horror Game
-
Anko's Loop is a short horror game where you play as a high school student trapped in her school. Something is after you, and you must find clues to prevent your deaths which reset a time loop. Will you uncover the story behind why are you trapped? This game was made in Unreal Engine.
- Star Cars
-
Throughout the course of CSE 420, our group created a racing game that takes place in space.
- Beacon
-
Beacon is a tower defense and farming simulator game created in Unreal Engine. You are the sole human being on a planet that is not your own. But be warned, during the night, the creatures of this planet will attack. Grow crops for materials then build towers to defend yourself. Keep your beacon intact, as that's your last hope of getting rescued.
- Golf Mania
-
Our game is about a arcade style golf game where each turn players are able to put obstacles that hinder other, and the goal is to get the hole with lowest strokes
- Decay
-
Horror/Puzzle game taking place in Chernobyl. Traverse around the area and to different abandoned buildings collecting key cards (or other items) in order to reach the reactor/escape.
(CSE 442: Software Engineering)
- HomeMate
-
- Jennifer Furdzik
- Shane Sworts
- Jillian Manjarrez Guerrero
- Aaron Radliff
- Yuktha Reddy Vanteru
- Dharmi Khadela
HomeMate makes subleasing simple. Browse verified short-term rentals, post your sublease in minutes, chat directly with potential tenants or roommates, and read honest reviews before committing. No more sketchy listings or stressful landlord negotiations—just a smooth, transparent subleasing experience.
- Recall
-
Plan, Reflect, and Relive with Recall “Capture your days with journals, moods, and voice notes—then rediscover your story through mood-based filters, an interactive calendar, and personalized insights. Recall turns moments into memories and memories into meaning.”
- Gatorade Zero
-
The ultimate online hub where players worldwide connect, compete, and enjoy gaming together.
- Auto Office Hours
-
Office hours scheduling app for Professors & TAs
- Ryan's
-
A web platform featuring interactive carnival and casino games like roulette, balloon pop, and blackjack, built with a focus on smooth gameplay and engaging UI.
- Margn
-
An online platform for readers to rate books, discover new titles, and participate in community discussions around shared interests.
- Pupperazzi
-
Our vision is to create the go-to platform for dog owners who want to enrich their pets’ lives through safe, fun, and meaningful social connections. By offering an intuitive swipe-based matchmaking system, we make it simple for owners to find compatible playmates based on personality, energy level, and lifestyle fit. The app fosters a trusted network of dog lovers who can easily schedule playdates, explore dog-friendly spaces, and watch their pets grow in confidence and happiness. Over time, this platform will become more than a scheduling tool — it will be a community hub where dogs thrive, owners connect, and lasting friendships are built on both ends of the leash.
- Whitespace
-
Our vision is to provide a simple, intuitive whiteboard tool that empowers individuals to capture, organize, and visualize their ideas in a way that feels natural and flexible. Whether it’s brainstorming for a project, planning an event, or mapping out study notes, the app makes it easy for one person to shape their thoughts and then share the board with friends for feedback and suggestions. By focusing on clarity and ease of use, the platform helps turn scattered ideas into structured plans, making creativity and organization more accessible to everyone. Over time, this whiteboard can grow into a shared space for group planning and inspiration, but its foundation will always be about helping people bring their ideas to life with simplicity and confidence.
- Gamblers Paradise
-
Online casino app that lets users bet and earn by watching ads.
- Lettuce Turnip the Beet
-
A nutrition application which makes user's lives easier by providing a simplistic method of finding, saving, and creating recipes, as well as sharing them with communities.
- Pro Gambler
-
A snazzy gambling website with 4 casino favorites!
- FlowCheck
-
FlowCheck is an end-to-end content approval platform designed to replace chaotic, email-driven review cycles with a structured, role-aware workflow. Our system centralizes asset upload, reviewer assignment, commenting, deadlines, and audit history into a single interface. Each user sees a personalized dashboard based on their role (Marketer, Medical, Legal, Regulatory), allowing them to track tasks, review files, and resolve threads without friction.
- Lemonwire
-
This project is a social music platform designed for avid listeners, casual fans, emerging artists, and music critics to connect through shared music tastes. The app emphasizes discussion, discovery, and expression rather than just consumption, enabling users to: Share and rate curated playlists, reviews, and listening activity. Discover new music through friends, mutual connections, groups, and a recommendation algorithm. Link Spotify/Apple Music accounts to showcase recently played tracks. Join themed groups (“rooms/stages”) to exchange music preferences and explore genre-based communities. Participate in features like weekly playlist competitions to add a fun, competitive edge. For emerging artists and critics, the app serves as a platform to build credibility, market themselves, and influence listening trends. Ultimately, it creates a dynamic, collaborative environment where music appreciation becomes a social experience, fostering friendships and communities around shared tastes and live events.
(CSE 455: Introduction to Pattern Recognition)
(CSE 546 Reinforcement Learning)
- A Multi-Agent Reinforcement Learning Framework for Autonomous Security & Intrusion Detection.
-
This project presents an AI-driven cybersecurity defense system that leverages Multi-Agent Reinforcement Learning (MARL) and Graph Neural Networks (GNNs) within the CybORG++ simulation environment. The framework models a digital network as a dynamic graph, where Red-Team (attackers) and Blue-Team (defenders) agents continuously adapt through adversarial learning.
(CSE 587: Data Intensive Computing)
- Big Data–Driven Vehicle Characterization Using Distributed ML Techniques
-
The project is focused on Car Sales Dataset by performing the data cleaning, exploratory data analysis and define the Machine Learning problems and objectives which are solved and providing with information insights. The primary objective of this project is to learn and understand the real world machine learning problems and solve them using PySpark and its libraries.
- DIC Buffalo Tree Dataset Analysis
-
Our group used the Buffalo tree inventory dataset to examine multiple Machine Learning questions about the dataset. These questions include mapping potential hot spots for CO2 to be sequestered, predicting trees with a high likelihood of blocking roads during emergencies, classifying tree species as high benefit and low benefit depending on characteristics, and predicting the kWh saved at current and potentially future tree sites.
- Data Intensive Computing Project
-
Big Data Analysis and Machine Learning Models on LinkedIn Jobs and Skills Dataset
- Car Sales Analysis Using Data Cleaning, EDA, and Spark-Based Machine Learning
-
This project focuses on analyzing a large used-car sales dataset to understand the factors that influence a car’s resale price and to build a Spark-based machine learning model capable of predicting it accurately. The workflow includes data cleaning, exploratory data analysis (EDA), feature engineering, and the development of a supervised regression model using PySpark’s MLlib. As part of the analysis, we examined key attributes such as brand, year of manufacture, fuel type, transmission, engine size, and mileage to identify their impact on pricing patterns. EDA techniques—including correlation analysis, boxplots, scatter plots, and trend visualizations—were used to uncover relationships and outliers in the dataset. After preprocessing the data with Spark pipelines (handling missing values, encoding categorical features, and assembling feature vectors), a Random Forest regression model was trained, evaluated, and improved through hyperparameter tuning. The final model achieved high predictive accuracy and provided insights into which features most strongly affect resale price, helping illustrate real-world pricing behavior in the used-car market.
- E-Commerce Insights and Predictive Data Analysis Using PySpark
-
This project explores how users behave on an e-commerce platform by analyzing a large dataset with PySpark. We cleaned and processed the data, studied browsing and purchase patterns, and built predictive models to understand what drives conversions. Our findings reveal clear user segments, strong brand preferences, and insights that can help improve customer engagement and boost sales.
(CSE 611: Master's Capstone Project)
- Youro Mobile
-
We developed a mobile app for the Youro personal health care platform, which connects patients directly with urologists for the treatment of seven common diagnoses.
- Safety Knights
-
- Bhavitha Bandla
- Bhavneet Kaur
- Lakshmi Sreya Kuppa
- Sai Shashi Kiran Rachabattuni
- Varun Teja Goud Madhagouni
We extended the existing Safety Knights platform by building four major modules to make the system more actionable for safety professionals: - A module for adding and managing safety school information. - An integration to view detailed OSHA records for various industries. - An interactive global map that visualizes the locations of professionals, schools, and OSHA-registered companies. - A comprehensive badging system where badges are awarded automatically based on user actions or assigned manually by administrators.
- Student Linter in OCaml
-
A student linter written in OCaml that checks their compiled code against a style guide and gives the student relevant feedback.
- SVC Central
-
This project is for developing a service marketplace platform connecting customers with local service providers that includes in-built scheduling and employee management.
- Clairvoyix
-
Clairvoyix is Advance2000’s next-generation private AI & GPU compute platform that makes high-performance computing as simple as using an app. It lets architects, engineers, analysts, and creators run GPU-accelerated AI workloads without any technical setup. Clairvoyix handles everything behind the scenes GPU allocation, containerization, model deployment, and performance tuning and gives users a clean, intuitive interface.
- Filmic Inspection Report
-
- Isha Nitin Shetye
- Anurag Rajesh Mahajan
- Atharva Nitin Prabhu
- Atharva Bhalchandra Wadekar
- Jaya Krishna Pasham
Filmic Technologies is the company, where the project is dedicated in helping preserve old films by automatically analyzing every frame of newly digitized reels. It captures details like damage, dirt, color fading, audio issues, and even edge codes, then turns all of that into clean, easy-to-read reports. This makes it much easier for archivists to understand a film’s condition and decide what needs restoration. Ultimately, Filmic helps protect cultural memory by giving fragile films a second life in the digital world.
(CSE-546: Reinforcement Learning)
- Adaptive RL Tutor for Algebra Mastery
-
Our plan is to create and check out an adaptive tutoring system that uses reinforcement learning (RL). Think of RL as a smart way to personalize how hard the next question is and what kind of feedback a student gets while they’re learning algebra. Most online learning tools right now follow really rigid, pre-set rules, like ”you get to move on after you answer two questions correctly.” That just doesn’t work for everyone. It can’t keep up with a student’s unique pace, how quickly they’re truly grasping a skill or even if they’re starting to get bored. Our RL-powered tutor, on the other hand, will pick problems and decide on hints on the fly, all with the goal of making sure students really remember what they learn.
(CSE587)
- DIC - Amazon Books Review
-
The datasets Amazon Book Reviews are used. Included the analysis, cleaning and exploratory analysis of the dataset.
- Yellow Network
-
Analyze NYC Yellow Taxi trip data using PySpark to handle large-scale processing efficiently. The dataset contains detailed information about each taxi trip, including timestamps, locations, distances, fares, and tips, which enables us to study patterns in urban mobility and passenger behavior across New York City. By cleaning and structuring the data, we prepared it for reliable analysis and modeling at scale.
- Alpha Core
-
This project established a scalable big data pipeline using HDFS to manage and preprocess approx 600,000 financial records (Phase I). We then implemented four machine learning models, including Random Forest and LSTM, to predict stock prices with high accuracy (R^2=0.983$) and forecast trends (MAPE 0.45%), while confirming the difficulty of directional classification (Phase II).
- CarDealers
-
This project builds a clean PySpark pipeline to predict a car’s mileage from dataset which includes age, engine size, fuel type, model information, and price. After rigorous data cleaning (deduplication, winsorizing extremes, and miles-per-year sanity checks) and light feature engineering (age², engine², log-price), we train a Gradient-Boosted Trees model with a temporal train test split and cross validation. Performance is summarized with RMSE and R², and we operationalize results by ranking vehicles via absolute residuals to create a “risk list.” This list highlights potential odometer or data-quality issues, focusing review on the most anomalous cases.
(EAS 563)
- Bolt-Unitree Go2
-
This project enables the Unitree Go 2 robot to autonomously find and move towards a target ball (green, pink, or yellow) using its onboard vision. The robot runs a custom-trained YOLOv8 model to detect and track the ball's position in real-time from its camera feed. This vision data is fed into a "visual servoing" control loop: Steering (Yaw): The loop calculates the ball's horizontal position (x coordinate) in the frame. If the ball is not in the center, it sends a yaw (turn) command to the robot to steer left or right until the ball is centered. Movement (Forward): The loop simultaneously checks the size of the ball's bounding box. If the ball is small (far away), it commands the robot to walk forward. As the robot gets closer, the ball appears larger, and the robot slows down, stopping when it reaches a target distance. This creates a complete, closed-loop perception-to-action system that allows the robot to "lock on" and autonomously follow its target.
(Master's Project (CSE 799, not CSE 611))
(Project Submission)
(CSE 4/587: Data Intensive Computing Project Demo)
(CSE 455 :course project )
- Detecting AI-Generated Images (Real vs FAKE) AI)
-
This project aims to detect whether an image is real or AI-generated by analyzing spatial textures and frequency-domain artifacts. The goal is to build a robust system that can reliably identify AI-generated images and understand the patterns that distinguish them from real photographs.
(CSE 587)
- Buffalo Tree Inventory Squad
-
We are trying to identify, solve and enhance of the total environmental and economical output of the trees in Buffalo city.
- Team eCommerce 587
-
eCommerce Behavior Data Exploration
(CSE 587LEC A Data Intensive Computing)
- eCommerce behavior data from multi category store.
-
Online retail platforms generate large volumes of user activity logs that contain valuable behavioural and product-level insights. The purpose of this project is to extract insights from a sample reality e-commerce dataset on user engagement, top performing categories, price and purchase behaviour. Leveraging Spark for scale and ML models for prediction, this work illustrates how data-driven guidelines contribute to a retail strategy and decision.
(CSE 676 - Deep Learning)
- Retrieval-Augmented Radiology Summarizer (RARS)
-
Radiology departments are experiencing growing workloads as the number of chest X-rays continues to rise, making rapid and accurate interpretation essential. This project introduces an AI-powered Radiology Chatbot that integrates retrieval-augmented generation (RAG), image captioning, and summarization to support clinical decision-making. The system processes DICOM chest X-ray images and free-text radiology reports from the Indiana University OpenI dataset, enabling multi-turn, context-aware conversations with clinicians. Using semantic embeddings and FAISS indexing, it retrieves the most relevant report sections and metadata to enrich query responses. A BART-based summarization module generates concise and clinically coherent report summaries for quick reference. The chatbot is deployed through a Streamlit interface that supports both chat-based Q&A and one-click summarization. Overall, this system aims to streamline radiology workflows, enhance interpretability, and improve accessibility of diagnostic insights for healthcare professionals.
(CSE587 - DIC)
- Yahoo Finance Analysis
-
This project demonstrates a full-cycle Big Data pipeline analyzing over 600,000 Yahoo Finance stock records. Transitioning from local Python analysis to a distributed Hadoop/Spark environment, we implemented scalable ML models to forecast prices (Gradient-Boosted Trees), classify market direction (Random Forest), and cluster companies by risk profile (K-Means). The results highlight the challenges of predicting market direction while showcasing the power of distributed computing for financial insight.
(CSE587: Data Intensive Computing)
- Data Analysis and Machine Learning on Yahoo Finance Dataset
-
We utilized PySpark to perform Data Cleanup, Exploratory Data Analysis (EDA), and Machine Learning Algorithms on a specific Yahoo Finance Dataset, keeping track of various factors such as anomaly detection.
(Independent Study / Research)
- Mesh Interpolation
-
An improved method for interpolating the intermediate points in a partial mesh of triangular cells.
- LipSinc(DNx)
-
My project focuses on developing a multi-modal speech denoising system that enhances speech clarity by combining both audio and visual information. Building on my work from CSE 676: Deep Learning, I utilize the CleanUNet audio backbone and integrate lip movement analysis using MediaPipe Face Mesh to provide visual context that helps separate speech from background noise. My goal from this project is to create a more intelligible speech enhancement model that performs even in noisy, real-world environments.
- Weather-informed imputation of block-missing highway speed data
-
This project applies generative diffusion models to impute missing traffic speed data using connected vehicle and weather information. It enhances data reliability for intelligent transportation systems and winter road safety analysis.
- Collaborative Truck–Multi-Drone Delivery Systems: Scheduling and En-Route Operational Optimization
-
This thesis examines a cooperative truck-drone delivery system, where multiple drones serve off-route clients while a truck follows a fixed path and acts as a mobile base.
- A Conditional Variational Autoencoder (cVAE) for Subject-level Concentration–Time Curve Reconstruction
-
In this project, we worked on predicting a patient’s drug concentration–time profile using deep generative models. After testing traditional methods like SVMs and mixed-effects models, we designed a conditional VAE that captures the nonlinear PK behavior missing in simpler models. The model not only produced far more accurate reconstructions but also learned interpretable latent patterns linked to patient covariates. This helped create a reliable tool for generating full PK curves with uncertainty estimates.
- Online Algorithms with capacity constraint
-
Minimise the maximum flow time for online algorithms.
- Using Data to Draw on CD
-
There have been attempts to draw on optical discs, specifically CD’s, using the data written to them. These attempts were recorded online but do not provide specifics on data computation, the methods of determining these processes, and an explanation of their final implementation. On top of this, when first using the implementations we were unable to produce the same results they yielded, or determine the constraints to produce the correct result. For this project, we determined how to draw on a CD using data and document the process so that it is reproducible, such that if someone in the future wants to build upon our findings they can. We did so by reverse engineering the existing implementations and contributing our own versions.
- Application Layer Security for Wi-Fi Sensing
-
IEEE 802.11bf is an amendment to the Wi-Fi standard that enables “Wi-Fi sensing” which allows devices to use Wi-Fi signals to sense information about the environment as opposed to the traditional use of transmitting data. Such applications include motion detection, presence sensing, gesture recognition, and more. It does this by utilizing channel state information to determine changes to the environment. Security and privacy is an open issue with this standard alongside being able to coexist with regular Wi-Fi communication. The aim for the project at this time is to showcase the research done this semester into potential solutions to privacy and security, discussing the performance, security, and ease of implementation of various methods.
- Implementing typed SQL support in Draupnir
-
Draupnir needs support for SQL for various issues. This goes through converting SQL into AGRA.
- Analysis of Single vs Double Precision Performance in Numerical Methods
-
Performance analysis of single vs. double precision formats in numerical pde solvers, measuring computational efficiency and solution accuracy.
- An Investigation of the Visibility and Correctness of Read-Only Transactions in Fault-Tolerant Systems
-
This research investigates which existing concurrency control protocols claim to support invisible read‑only transactions, and how their correctness and performance hold up in replicated, fault‑tolerant settings.
- craft4free
-
Craft4Free is a cloud-native, distributed Minecraft server hosting platform that lets users launch and manage Minecraft servers on cloud instances. Instead of forcing users to handle complex setup or infrastructure, Craft4Free dynamically provisions isolated worker instances, deploys optimized server environments, and continuously monitors performance and demand. Its controller scales resources up or down in real time to optimize cost and share resources between servers. The site consists of a React frontend with a distributed Golang backend consisting for a controller and workers. It uses Redis for distributed communication.
(CSE 546: Reinforcement Learning)
- CityLearn-DR: A Multi-Hazard RL Benchmark for Disaster-Ready Grid Control
-
Modern reinforcement learning benchmarks for building energy management assume stable operating conditions, even though real distribution systems face rare but severe shocks from climate extremes and malicious grid disruptions. This gap raises a fundamental question: can an RL controller learn resilience- aware behavior when exposed to stochastic multi-hazard disturbances with advance warning signals. We address this by introducing a disaster-augmented variant of the CityLearn Challenge 2022 Phase 1 environment that models snowstorms, heatwaves, wildfires, hurricanes, and grid attacks through dynamic modifications to load, solar generation, and grid availability. The state representation is extended with hazard type, warning indicators, and time-to-event features, and we train a centralized Proximal Policy Optimization (PPO) controller using an urgency- weighted objective that prioritizes continuity of supply during hazard periods. In one-year rollouts containing three major disaster events totaling 288 hazard- hours (72-hour winter storm, 120-hour heatwave, 96-hour wildfire), the learned policy maintains a long-run average battery state of charge of 51.8 percent despite sustained discharge during outages, raises SOC proactively in advance of hazard windows, and limits unserved energy to 3.42 kWh per disaster-hour, outperforming random and rule-based baselines while preserving peak-demand reductions. These results show that integrating disaster forecasts directly into the control loop enables RL controllers to internalize temporal risk structure and develop proactive resilience strategies rather than reactive responses.
(CSE587 - Data Intensive Computing)
- Customer Behaviour Analysis and Recommendation
-
This project focuses on building a complete end-to-end big data analytics pipeline using Hadoop for large-scale data ingestion and PySpark for distributed data processing, exploratory analysis, and machine learning. Using the eCommerce Behavior Data from a Multi-Category Store dataset, which contains over 110 million user interaction events, the goal is to understand customer behavioral patterns, estimate revenue trends, predict purchase likelihood, segment customers, and identify cross-category product affinities.