2025 Computer Science Capstone Symposium

The Computer Science Department senior capstone presentations will take place Friday, May 9th and Saturday, May 10th in Xavier 201 on the Pacific Lutheran University campus.  If you’d like to join the capstone Zoom session, please email Assistant Professor Jeff Caley at caleyjb@plu.edu.

Friday, May 9th

2pm – Vehicle Detection and 3D Pose Estimation
Braden Brooker (BS), Hailey Ledenko (BS)

Vehicle detection and 3D pose estimation are critical challenges for ensuring safety in autonomous driving. While companies like Waymo have demonstrated the potential for safe autonomy, their reliance on expensive and resource-intensive depth-sensing technologies such as LiDAR limits scalability and accessibility. Our work addresses these cost and computational barriers by relying solely on monocular camera data to estimate the 3D position and orientation of vehicles—similar to Tesla’s vision-based autonomy approach. At the core of our system is EgoNet, a deep learning model that combines heatmap regression for 2D vehicle keypoint detection with a fully connected network that lifts these key points into 3D space. The resulting spatial data is processed using ROS and visualized within a simulation environment. EgoNet’s key innovation lies in its ability to produce accurate 3D pose estimates, even for vehicles that are partially occluded—a challenging edge case in which many models struggle. Upon its publication at CVPR 2021, EgoNet outperformed prior state-of-the-art methods when tested on the KITTI dataset, a widely recognized benchmark for autonomous vehicle perception. Our work leverages EgoNet’s high precision 3D spatial inferences and uses these model output to render a 3D simulation of RGB keyframes using RViz. These successful reconstructions provide a strong foundation for real-time obstacle avoidance systems, moving us closer to scalable and safe autonomy without the need for expensive sensor suites.

2:30pm – OptiMiX
Yuta Shimazu, Data Science (BS)
This study proposes an integrated marketing mix modeling (MMM) that melds time series analysis and machine learning toward measuring media channel effectiveness. The integrated model “OptiMiX” includes Prophet for baseline sales forecasting, Random Forest for estimating non-linear effects of media, and SHAP values for explanation. Transforming media spending with geometric Adstock lets the model estimate carryover effects. The framework also considered spending efficiency and marginal returns channels, analyzing in a spend-effect share context. The analysis results show that my model successfully separated seasonality and outside factors from the impact of real media and produced valuable insights for budget optimization. The model was validated with Meta’s MMM dataset, where the framework achieved <10% prediction error (RMSE/MAE in relation to revenue) and found around 2% decremental error was available via Adstock parameter tuning. This model also balances the statistical analysis requirements while remaining usable at an operational level.

2:45-3:15pm – Break

3:15pm – HoopIntel
Stacie Spahr (BA), Ashley Akamine (BS)

HoopIntel is a machine learning-powered application aimed at automating basketball game analysis to support scouting and coaching decisions. Our project focused on developing a pipeline to detect the court, identify players, and track their movement across frames using object detection and re-identification models. While we were not able to fully complete the end-to-end system for generating scouting reports, we successfully trained and tested individual components, including a YOLO-NAS object detection model and a MobileNetV3-based Re-ID model with Faiss similarity search. These models lay the groundwork for a scalable solution that can reduce the time coaches spend manually reviewing film and provide insights into player tendencies and team strategies.

3:45pm – Enhancing Quality of Service (QoS) Monitoring for Smart Tracking Devices
Clara Gordon (BA), Yu-Ri Hahn (BS)

This project, developed with Codepoint Technologies, focuses on improving Quality of Service (QoS) monitoring for smart tracking devices used in asset management. These devices generate vast amounts of data, with battery health being a critical factor affecting system reliability. We developed a custom API plugin to automate data entry into a Docker-managed database and created a function to calculate key battery performance metrics, such as rate of change. The processed data is visualized using the Google Charts API, allowing users to track performance over time. Throughout the development process, we collaborated closely with the technical lead to ensure alignment with the project’s objectives. The project underscores the importance of battery health monitoring in real-time tracking systems and lays the foundation for predictive maintenance and more efficient asset management.

Saturday, May 10th

10am – Fredrick
Roderick Percival (BA)

Fredrick is a discord bot and website combo that puts multiple server’s messages in one database, allowing for searching across multiple servers at once. This search can be done on any server that the discord bot is connected to, and all the messages from that respective server are available for view from any other connected server. Utilizing discord’s slash command feature, the user can see what parameters are available for use, and it displays instructions of how the parameters are intended to be used. The website shows a live feed of all of these messages with the same level of control as the discord slash command implementation, but without the limitations of discord’s client, allowing for a more pleasant viewing experience that can also be filtered. Both are focused on being simple to use, allowing someone with no experience with database languages to query with modular prompts that can be added and removed with an easy to use interface.

10:30am – Through Science Comes Art!
G. Alvarado (BA)
Welcome all young and old to the future of movie magic! 2D animation remains a powerful storytelling medium, yet its resource-intensive nature has made it increasingly rare in today’s industry. What if we could change that? Could A.I work with, rather than against artists, making 2D animation more accessible? Can a small studio implement this and revive this beloved genre? Join international award-winning filmmaker G Alvarado as we explore cutting-edge image generation and video interpolation A.I models. Along with an enhanced 2D animation pipeline that preserves artistic integrity using customly trained models. Early findings suggest that this can significantly reduce production time, transforming what once took years into mere months. Come all far and near to peek behind the curtain. For once you do, you will find through science comes art, and through innovation, a new era of storytelling begins!

10:45-11:15am – Break

11:15am – Roko’s Basilisk
Evan Archer (BS), Zane Davenport (BA), Christopher Brice (BA), Wasif Ramzan (BS)

Roko’s Basilisk is a roguelike first-person shooter game that seeks to redefine player immersion by offering an adaptive, procedurally generated experience tailored to individual playstyles. Through the use of a custom rule-based AI we’ve created, the game employs constraint satisfaction techniques to analyze a number of performance statistics gathered at the end of each level. These statistics are then processed to identify player strengths and weaknesses. This analysis guides the dynamic construction of subsequent levels, ensuring that each level challenges and evolves alongside the player. In conjunction with this, we have also created a behavior tree to guide the enemy combatants the player will be facing. These enemies will also have their own stats adjusted in accordance to the player’s performance throughout the game. The result is a uniquely personalized gameplay experience where no two players, nor gameplay sessions, are ever identical. Roko’s Basilisk not only entertains but also explores the cutting-edge potential of artificial intelligence in game design.

11:45am – Votify
Eric Sun (BS), Vincent Allen Sison (BS), Andrew Sandoval (BS)
Votify is a social music application designed to enhance group interaction and bonding through collaborative playlist creation. Leveraging the Spotify API, it allows users to create “Votelists” — dynamic, shared playlists where participants can vote on which songs are added. Built with Flutter and Dart for the frontend, Node.js for the backend, and PostgreSQL for data management, Votify delivers a seamless and engaging user experience that combines music with social engagement.

12:15-1pm – Lunch

1pm –  Zenith Racers
Brody Willard (BS), Emmanuel Obikwelu (BS), Robert Marti (BS)

Zenith Racers is a high-speed, anti-gravity racing game developed using Unity. The project focuses on delivering an engaging experience with unique gameplay mechanics, such as vehicles defying gravity and clinging to 3D tracks, alongside features like customizable difficulty levels and multiplayer support. This game integrates Unity’s physics engine and Blender-designed assets to craft visually stunning tracks and vehicles. The goal is to provide an intuitive, easy-to-learn experience that is challenging to master, catering to both casual and competitive players.

1:30pm – GildBudget
Luke Coleman (BS), Ryan Mitchell (BS)
GildBudget is a full-stack web application developed to facilitate personal financial management through intuitive budgeting, expense tracking, and account aggregation. Utilizing a React-based frontend and an Express/Node.js backend, the application integrates with the Plaid API to securely import and categorize financial transactions from user-linked bank accounts. Transactional and account data are stored in an SQL database, enabling efficient querying and analysis. GildBudget provides users with dynamic visualizations and budgeting tools to monitor spending patterns, track financial goals, and make data-informed decisions. The system emphasizes usability, data privacy, and extensibility, serving as a robust platform for individual financial literacy and long-term planning.