2025 ECE Design and Innovation Day

Project Description: AV1 is a popular open-source video compressor, but when compressing a large volume of videos, it may consume a tremendous amount of power. Our capstone project uses artificial intelligence to decrease the compute necessary in AV1 video compression without losing quality. By integrating neural networks, we make the process more efficient, which helps streaming services, video creators, and anyone who needs high-quality videos with smaller file sizes.

Project Description: Modern data centres use an enormous amount of energy to coll their vast fields of server racks. At the same time, residential and office buildings use energy to heat up their interiors. Our project, sponsored by Professor Christoph Sielmann at UBC, consists of a feasibility investigation that explores the ability for computers to take on the role of heating buildings. Using the heat from running servers to heat buildings has the potential to remove two points of energy consumption and decrease the carbon footprint of data centres and the heating of our homes.

Through this project, we have enabled the data centre to “spread out” across many locations, each with their own compute unit. Job computations do not have to happen in a central data centre anymore, and can now happen in office buildings, schools, and basements. In addition, we have also proved that computers do have the ability to heat enclosed environments, and more powerful hardware has the potential to heat whole buildings.

Sam Dai daisam1215@gmail.com; Nancy Makar nancynagy66@gmail.com; Jomari Francisco fjomari100@gmail.com

Project Description: This project aims to develop a carbon capture data monitoring website and an immutable carbon accounting system for Mitico Tech. By integrating real-time IoT data collection with blockchain technology, the system ensures transparency and accountability in CO₂ tracking. The platform leverages MQTT for real-time sensor data transmission from up to 1000 Arduino devices, with Eclipse Mosquitto as the broker, securely deployed with TLS encryption. A Cloud SQL database stores validated sensor data, while a web application provides interactive dashboards for both internal users and public stakeholders. 

A major technical achievement of this project is the combination of an efficient MQTT-based data pipeline with a blockchain-backed carbon accounting system. The MQTT system enables high-frequency, real-time data ingestion, with subscribers processing and securely storing data in a cloud-hosted SQL database. Meanwhile, Hyperledger Besu, an Ethereum-based private blockchain, ensures immutable CO₂ tracking through smart contracts. By utilizing a Proof of Authority (IBFT 2.0) consensus mechanism, the blockchain achieves high transaction throughput (~2000 TPS) and immediate finality. Internal users can securely log CO₂ transactions, while external stakeholders have verifiable, read-only access. This integrated design ensures that the platform remains tamper-proof, scalable, and cost-effective, requiring minimal ongoing maintenance.

Project Client: Rhino Ventures

Project Description: One of the biggest challenges in adopting lighter-than-air vehicles (LAVs) for commercial use is the lack of landing infrastructure, such as masts or landing areas. Our project aims to develop an autonomous control system that enables LAVs to maintain their position with minimal energy consumption and without the need for ground infrastructure. 

To achieve this, we will use multiple regression methods to analyze collected flight data from the Inertial Measurement Unit (IMU) and estimate the blimp’s state-space model. Based on this model, our control system will implement a Linear Quadratic Regulator (LQR) to achieve optimal control. This approach ensures the blimp can autonomously stabilize itself in real-time under varying environmental conditions, holding position anywhere during operation without reliance on ground infrastructure.

Sahil Lakhani sahillakhani2002@gmail.com; James Chan sfxcjc@student.ubc.ca; James Lin jlin001015@gmail.com; Jiajie Sheng sjj2021@student.ubc.ca

Project Description: Our project aims to streamline and automate startup evaluations using LLM. The problem it solves is the time-consuming and inconsistent nature of manual startup assessments, which often depend on subjective opinions and fragmented data.

It benefits investors or business consultants who need a structured, data-driven way to assess early-stage startups, as well as Startup founders looking for clear, actionable insights on how to advance their business.

A major challenge we tackled was balancing model performance with cost-effective processing. Evaluating startup data requires analyzing complex information, which can be computationally expensive if processed inefficiently. Initially, our system attempted to process entire datasets at once, leading to high token usage and increased computational costs. To overcome this, we redesigned our approach by integrating a context-aware retrieval system using ChromaDB. Instead of feeding the entire dataset to the AI model, our system retrieves only the most relevant information, significantly reducing processing load while maintaining accuracy.

Project Description: PURE-SIM is a decision-support tool designed to streamline the feasibility assessment of Run-of-River (RoR) hydroelectric projects. Engineers often face significant challenges in evaluating potential RoR sites due to the complexity of data collection, analysis, and cost estimation. PURE-SIM automates this process, providing data-driven insights to support decision-making.

Our solution integrates site-specific hydrological data with power generation modelling to estimate energy output and financial feasibility. This reduces the time, resources, and potential human errors involved in traditional feasibility studies.

Zijia Wang zijiawng@gmail.com; Mehmet Berke Karadayi mberkekaradayi@gmail.com; Son Nguyen nguyenteo2u@gmail.com; Chayanika Awasthi chayawa@student.ubc.ca 

Project Description: Many BC Hydro customers in rural, remote, and reserve areas do not have an address that follows Canadian addressing standards. The current outage reporting system, primarily via phone, makes it difficult for BC Hydro to accurately locate and resolve outages, leading to long wait times for customers. Our solution is a full-stack web app that enables these customers to outline the affected area on a map and submit outage reports directly through the app, improving communication and response time. The app features a custom-built map interface that integrates multiple data sources to accurately represent rural and remote regions.

For further details, please reach out via LinkedIn: Maggie Gu (@maggieh-gu); Aleksandra Vujicic (@aleksandra-vujicic); Evelyn Sankar (@evelynsankar); Hod Kimhi (@hodkimhi); Bar Nyhof (@barnyhof). Via GitHub: Natalie Balashov (@natmarbal).

Project Description: The aim of our project is to build a Brain-Computer Interface (BCI) device that will allow users to control applications or devices. Effectively, this project provides an alternative way for people to interact with computers, as opposed to a traditional mouse and keyboard. The target audience of our project includes general consumers, however those with trouble moving their hands (e.g.: paralysis patients) would likely benefit the most.

Major Design Contributions:
1. Developing a PCB capable of measuring bio-signals, e.g.: Electrooculography (EOG) and Electroencephalography (EEG), and transmitting measurements over Bluetooth.
2. Developing software models capable of extracting commands (or user “intent”) from measured bio-signals.
3. Creating a user-friendly interface. This includes both software that enables users to configure and practice using their devices, and physical cases that allow users to comfortably wear their devices.

Charles Surianto charlessurianto@gmail.com; Yifeng Liu laveee.liu@gmail.com; Jerry Ge jerry.jge@gmail.com; Kerry Wang kerrywang369@gmail.com; Ben Leitch benleitch17@gmail.com

Project Description: The purpose of our project is to create a software solution for an indoor automated hydroponic garden that will promote sustainable food production among people living in urban environments. Our solution addresses key shortcomings in the current market for hydroponic gardens by providing a user friendly mobile app and custom garden controls to make growing your own food easy and accessible. Our group has supported our client by providing a programming foundation upon which they can continue to develop their indoor automated garden.

Specifically, we delivered code frameworks for the mobile application, custom server, database, and hardware controller which are scalable and modular for any future capstone groups to work on. We made the user interface intuitive to operate and we made it easy for our client to adapt our solution to their eventual physical garden design. Because the client’s garden does not yet exist, we were able to demonstrate the functionality of the solution with the simulation of a hydroponic garden environment. 

Steven Yan steven.yan42@gmail.com; Wangchen Xu xuwangchen2003@gmail.com; Owen Lin owen.tl@outlook.com; Will Kenna wmkenna34@gmail.com; David McPherson davidharrymcpherson@gmail.com

Project description: We developed a scalable camera system to improve security and research monitoring for the Foundation of the Energy Collective (FEC), covering both indoor and outdoor spaces on their 130-acre property. Our system integrates into FEC’s existing infrastructure, providing live monitoring and recorded footage access to help prevent vandalism, track ongoing projects, and ensure the safety of their facilities.

Our system supports up to 100 cameras and delivers high-quality 1080p video at 30 frames per second with real-time viewing and minimal delay. Recordings are securely stored and updated every minute, ensuring footage is always available when needed.

We designed a simple and organized interface that allows users to easily access live streams and past recordings that is compatible with various different types of cameras. With different access levels for staff, managers, and administrators, we ensure that the right people have the right permissions. Our system also allows cameras to be added, managed, and configured to meet changing needs.

Etienne Gagnon etn.gagnon@gmail.com; Ryan Nedjabat ryannedjabat17@gmail.com; Julian Wong julian.connor.wong@gmail.com; Zheqing Yan yan_zheqing@163.com; Yufan Zhou F.Yufan@hotmail.com

Project Client: Orca Advanced Materials Inc.

Project Description: OrcaProbe is a reconfigurable thin-film probing system designed to accelerate the characterization of printed thin-film devices. It enables 2, 3, and 4-probe electrical measurements, automating testing procedures and reducing testing time compared to conventional methods. The system integrates a microcontroller-driven measurement platform, a flexible switching network, and a Python-based GUI for seamless data acquisition and visualization. While not a new invention, OrcaProbe was designed and built entirely from scratch to ensure full ownership for the client, allowing future modifications and scalability as their needs evolve.

Major Design Contribution: A key technical challenge we resolved was developing a fully reconfigurable probing system that supports multiple electrical measurement configurations while maintaining signal integrity and minimizing parasitic effects. We achieved this through a hybrid relay-based switching network, optimized embedded firmware for precise data synchronization, and a GUI that dynamically adapts to different measurement types. These design choices ensure accurate and repeatable measurements while providing a user-friendly experience.

Aaron Loh aaron.loh.aaron@gmail.com; Dipak Shrestha dpakbstha@gmail.com; Idil Bil idil.bil451@gmail.com; Kerem Oktay keremoktay121@gmail.com; Peggy Yuan pyuanvan@gmail.com

Project Client: UBC Surgical Technologies Lab

Project Description: Our project, in collaboration with the ISTAR group at Vancouver General Hospital, focuses on improving computer-assisted surgery (CAS) for mandibular reconstruction by integrating computer vision-based tracking. Traditional infrared (IR) camera tracking systems like the NDI Polaris rely on reflective markers, which can be bulky, expensive, and prone to occlusion when surgical tools or personnel obstruct the line of sight. To address these limitations, we developed LentiMarks. This visual marker-based tracking system uses a high-resolution Basler camera and computer vision algorithms for real-time, high-precision tracking.

We integrated this system with 3D Slicer, an open-source medical imaging and modeling software widely used for surgical planning. Our implementation involved developing a custom OpenIGTLink communication pipeline, allowing our computer vision algorithm to transmit real-time tracking data into 3D Slicer’s surgical modeling interface. This enables surgeons to visualize and track the patient’s anatomy and surgical tools with high accuracy. Additionally, we introduced a dual-tracking mode, allowing users to switch between LentiMark tracking and traditional IR tracking, ensuring flexibility and compatibility with existing surgical workflows.

By replacing bulky IR markers with compact, vision-based tracking, our system improves surgical precision while reducing cost and obtrusiveness. This makes advanced tracking technology more accessible, enhancing workflow efficiency and accuracy in reconstructive surgery.

Nicholas Amy nicholastamy@gmail.com; Leon Guo leonguo736@gmail.com; Kevin Jiang kevinjiang2313@gmail.com; Seiya Nozawa-Temchenko seiyant01@gmail.com; Sikher Sinha sikherjs0105@gmail.com 

Project Description: Fall-related injuries, especially among the elderly, have devastating consequences that can irrevocably alter an individual’s life. An inflatable vest is a product on the market that protects the user from a fall. However, online reviews for the currently available inflatable vests are poor, citing unaffordable retail prices, false positive system deployment, and poor manufacturing quality. Founded by Dr. Jimmy Wang, Sapphire Sky Solutions aims to develop a fall-detection airbag vest that is reliable, fashionable, and comfortable while addressing the deficiencies observed in competing products.

Major design contribution: For the software part of the project, the main technical challenge was to fine tune the parameters and find the perfect tradeoff so that the algorithm can not only accurately detect a fall but also minimize the false positives. To solve this challenge, we used an iterative design method, learning from the deficiencies of each iteration of the algorithm and constantly making it better.

For the hardware part of the project, the main technical challenge was to design a small, light, and energy-efficient hardware system that is easy and comfortable to wear and use. To achieve this, detailed circuit and PCB design is conducted, focusing on energy consumption and carefully selecting components for optimizing PCB size, weight and efficiency.

Duc Thang Huynh thangpro@student.ubc.ca; Pengyu Ji jipy03@student.ubc.ca; Mathew Wu mathewwu@student.ubc.ca; Tigran Hakobyan tik2003@student.ubc.ca; Adam Legros alegros@student.ubc.ca

Project Description: The objective of this project is to develop a portable, wireless ultrasound imaging system for NeuroPrior AI, a company dedicated to integrating artificial intelligence with medical devices to enhance healthcare accessibility, particularly in remote regions. Their mission is to leverage advanced AI-driven solutions to bridge gaps in healthcare infrastructure and improve diagnostic capabilities in areas where conventional medical imaging is either unavailable or prohibitively expensive.

To achieve this goal, our solution incorporates an off-the-shelf non-destructive ultrasound transducer alongside custom-designed proprietary hardware and firmware. This integrated system is engineered to enable real-time data acquisition, processing, and wireless transmission to a host device. The processed ultrasound images are then displayed using a custom-designed graphical user interface (GUI), offering an intuitive and user-friendly experience for medical professionals and field practitioners.

By providing a more affordable and portable medical imaging system, this project aims to contribute to improving healthcare accessibility. It aligns directly with NeuroPrior AI’s vision of combining AI and medical technology to democratize healthcare, ensuring that critical diagnostic tools are available to communities regardless of their geographic or economic limitations. This project could pave the way for further advancements in affordable AI-assisted medical imaging, early disease detection, and more efficient telemedicine solutions worldwide.

Nischay Joshi nischayjoshi2312@gmail.com; Amaan Memon amaanmmn@gmail.com; Midora Shiu midorashiu@gmail.com; Michael Tan mmichaeltann@gmail.com; Ricky Zang zangruiqi@163.com

Project Client: Gordon Lab, UBC Zoology Lab

Project Description: Gordon Lab is dedicated to understanding and exploring the neuroscience of animal feeding behaviors, focusing on Drosophila melanogaster (fruit flies) due to their small size, short lifespan, and simpler neural systems. They developed the Fly Proboscis and Activity Detector (FlyPAD) to track fruit fly interactions with food and paired it with the Sip-Triggered Optogenetic Behavior Enclosure (STROBE) to manipulate neural activity. By flashing high-power LEDs based on food choices, STROBE can influence the flies’ preferences, helping researchers understand how light stimulation affects feeding behavior.

Major design contribution: The FlyPAD and STROBE systems faced several challenges. FlyPAD’s arenas degraded over time and missed ~40% of fruit fly interactions. To improve durability and accuracy, we redesigned the PCB with a four-layer structure to reduce signal interference and increase the SNR. Additionally, FlyPAD could only detect interactions, not actual food consumption, so we developed an additional parallel plate capacitance sensor system to measure food consumption intake.

STROBE lacked customization, with fixed LED power and duty cycles, limiting neural stimulation experiments. We redesigned the module to allow researchers to adjust these parameters and added a control system that would autonomously optimize settings (variety of power and duty cycles) to maximize fruit fly interaction with food.

Alvina Gakhokidze alvina.gakhokidze@gmail.com; Harsh Gandhi hpgandhi1@gmail.com; Bryan Zhang bryan.zhang5168@gmail.com; Alex Xiong alexxiong28@gmail.com; Umair Mazhar umair.mazhar786@gmail.com

Project Description: Medical diagnostic testing is a cornerstone of modern medical care, providing accurate information about the functioning of many different organ systems to provide insight into disease processes and treatment progress. However, such testing requires full-featured laboratories to carry out, requiring hundreds of thousands of dollars in equipment investment. Many places in the world do not have such resources, and thus patients in those areas miss out on such lifesaving information.

Silicon photonic biosensors have great promise in providing point-of-care medical diagnostic tests to those in need at a low cost. The UBC SoC Lab has set out to further develop this technology, and has proven that the concept works. However, their existing test setups require bulky laboratory photonics equipment to take accurate measurements of analytes in samples.

Our project is a self-contained evaluation system for silicon photonic biosensors. This consists of a low-cost wavelength-tunable laser source, a temperature controller for the temperature-sensitive silicon photonic chip, photonic receiver hardware to receive photonic signals from the photonic chip, and a software system to capture test data and allow researchers to execute different test sequences. All of the hardware is integrated onto one compact PCB, allowing for increased portability and reducing cost.

Our project serves as an easy-to-use hardware and software platform that researchers can use to further evaluate these biosensors. It also serves as a bridge to a portable, point-of-care device that could be taken to market.

Bennett Galamaga bennettgalamaga@gmail.com; Callum O’Riley callumchristopheroriley@gmail.com; James Marx jamesrmarx@gmail.com; Peter van den Doel petervandendoel24@gmail.com; Suhail Khalil suhailkhalil2002@gmail.com; Tenna Yuan annet6280@gmail.com

Project Description: Our Project, Aisdom is a GPU hosting platform targeted towards academia. The GPU hardware required for coursework and AI research can often be costly or unavailable to purchase. Our platform aims to address this issue by creating a GPU hosting platform that academics can rent and interact with GPUs through a web application.

Major design contribution: The major challenge addressed in our project is setting up intuitive desktop environments that allows users to connect with available GPUs and remotely work with GPU-accelerated applications with minimized latency.

Jonathan Cao caojiasen@gmail.com; David Deng davidcdeng@gmail.com; Syed Araf Imam syedarafimam27@gmail.com; Varun Seshadri varunseshadri13@gmail.com; Kelvin Wong kelvin@kelvinw.com 

Project Client: Scoop Robotix

Project Description: Scoop Robotix provides digital automation solutions that help solar energy companies streamline operations, manage workflows, and track project progress. Their platform enables businesses to optimize installation processes, monitor team activities, and ensure compliance with project timelines. However, their customer success team currently relies on manual tracking to monitor customer engagement, which makes it difficult to identify inactive users or detect accounts exceeding subscription limits. This approach is time-consuming, prone to human error, and lacks predictive insights, limiting the company’s ability to proactively address customer needs.

SkyWatch automates this by integrating with Scoop Robotix’s infrastructure to analyze customer activity data. Using a transformer-based machine learning model deployed on AWS, it predicts platform usage trends, identifies disengagement risks, and detects over-subscribed accounts. The system stores insights in a database and presents them through a user-friendly web application, allowing the customer success team to take proactive measures.

By transitioning from manual tracking to AI-driven monitoring, SkyWatch enables Scoop Robotix to improve customer retention, optimize subscription management, and enhance overall engagement efficiency.

Project Client: Nur Al Huda Educational Initiative

Project Description: Nur Al Huda is an innovative AI-powered educational platform designed to enhance accessibility to personalized educational content through integrated hardware and software solutions. Recognizing that many elementary schools prohibit personal electronic devices, our project specifically addresses this limitation by developing a portable, interactive device approved for educational use within school environments. Utilizing advanced Natural Language Processing and the Nur Al Huda AI model, our design ensures seamless and secure access to engaging educational experiences tailored to student needs, while fully complying with school technology guidelines.

Mohammed Abdul Jabbar mohoabja@gmail.com; Antonio Qiao guanhua.qiao2020@gmail.com; Yue Li olivly1229@gmail.com; Aden Jabbar adenjabbar5@gmail.com; Fatima Jabbar fatima.jabbar5@gmail.com

Project Client: IRDI System

Project Description: Driven by the need to reduce greenhouse gas emissions, the hydrogen vehicle market is projected to grow 35.4% annually. However, a major challenge remains: current fueling systems rely on one-way communication or none at all. Vehicles can send data to fueling stations, but stations can not respond, limiting fueling efficiency.

Using Near Field Communication, our team NF Connect has developed a solution for this challenge. Similar to Apple Pay, Near Field Communication uses radio waves to enable real-time, two-way communication between the vehicle and the fueling station.This allows for precise fuel management, automatic pressure adjustments, and early issue detection – ensuring a safer and more efficient fueling process.

With our custom nozzle prototype, our solution is designed and tested to work with existing hydrogen fueling stations, making adoption quick and easy. By introducing two-way communication to the hydrogen fueling process, we’re solving a critical problem and helping to drive the future of clean energy forward!

Cheyenne Tu cheyennetu1314@gmail.com; Josh Lim  jawshl331@gmail.com; Frank Jin frankjin1302@gmail.com; Christophorus Hansen hansenkw@yahoo.com; Alex Martin alexmartin250@hotmail.ca

Project Description: Educational institutions often utilize multiple platforms such as Canvas, Piazza, and course-specific websites, resulting in fragmented course information and increased workload for instructors to maintain consistency across platforms. Issues such as long response times to student queries and the potential for inaccurate answers when relying on peer responses can hinder student performance and impact the quality of support.

Our AI assistant is designed to help students find answers to their course-specific questions quickly and accurately. It offers study tips, summarizes key concepts, and recommends valuable resources, enabling students to engage more effectively with their coursework. At the same time, it provides instructors with meaningful insights to enhance course delivery without requiring additional data management across various platforms.

Emphasizing advanced cloud-based technologies, this project focuses on delivering a cost-effective and scalable solution through AWS and LTI. By developing robust cloud formation templates, our initiative aims to simplify the integration of this AI assistant into existing Canvas environments, making adoption seamless for other institutions.

Project Description: This project is in collaboration with the team at the UBC Robotics and Control Laboratory, which is developing a ground-breaking system for conducting remote ultrasounds, especially in rural communities. As these communities are often far from where ultrasound experts and sonographers work, conducting in-person ultrasounds introduces travel overhead, which incurs delay that can hinder essential and timely healthcare. This system helps to reduce wait times for obtaining ultrasounds by allowing an expert sonographer to guide any novice through the ultrasound process using a mixed reality headset in real time over a wireless connection.

Our team was responsible for exploring different methods of improving two features of the current system – namely the generated 3D patient mesh and the introduction of a VR headset to improve overall depth perception. These are crucial to the sonographer’s experience when interacting with the system during an ultrasound.

Project Description: Eagle Eyes has partnered with us to integrate advanced AI capabilities into their prototype application, MapChat. This collaboration empowers emergency responders and planners to tackle crises like wildfires and floods more effectively through a Geographical and Situation-Aware Large Language Model (LLM) System. It addresses a critical challenge: to deliver critical information quickly to those who need it most. Our solution overcomes three key limitations of general-purpose LLMs: (1) the inability to interpret or interact with data and objects on a digital map, (2) a lack of domain-specific expertise for scenarios requiring specialized professional knowledge, and (3) limited capacity to maintain accurate contextual memory during prolonged operations.

With these enhancements, first responders can ask plain-language questions like, “Show me safety zones near the fire perimeter with first aid resources,” and receive precise answers with citations and real-time map visualizations. This enables personnel of all technical skill levels—not just GIS experts—to access critical insights swiftly and confidently, enhancing situational awareness and decision-making during emergencies.

Here’s how MapChat AI delivers real value through its key features:

Project Description: Deng Audio Research is an academic research company that specializes in the field of audiology, the study of audio and hearing. Recently they developed a groundbreaking technique for measuring the auditory response of the human ear. This technique involves multiple microphones placed along an audio impedance tube, with a human ear positioned at one end and an audio source at the other. The captured microphone signals are fed through a signal processing algorithm to approximate the reflectance of the human ear across the auditory spectrum. This type of measurement is an essential prerequisite for the prescription of hearing aids and could lead to widespread headphone calibration tailored to individual ears, which would represent a significant step toward hearing loss prevention.

Deng Audio’s technique improves upon similar existing methods due to its enhanced noise resilience, expansion to cover the entire human hearing spectrum, and cost-effective and efficient data collection. Although this method has been validated in a lab environment, no working prototypes exist that could lead to real-world scalability. Previous iterations iHear1 and iHear2 were unsuccessful in reaching full system integration due to various technical challenges.

Our project, the iHear2+ is the first fully integrated working prototype for Deng Audio’s method. It includes an audio tube with 5 microphones placed for optimal computational noise resiliency. The signals from the microphones are transmitted along high-quality cables to a motherboard that streams all five channels in parallel using analog audio converter chips and an FPGA-based streaming model. The data is reliably streamed over USB to a PC app where our DSP implementation of Deng Audio’s algorithm processes results for storage and display.

In summary, the iHear2+ bridges the gap between theoretical research and practical implementation, paving the way for real-world applications in audiology and consumer audio calibration.

Luke Matson lukematson39@gmail.com; Holly MacGillivray hmacg18@gmail.com; Henry Dong henry_dhc@yahoo.com; Jaydon Alexis jaydonjalexis@gmail.com

Project Description: The exponential growth of satellite deployments has created a critical challenge: orbital congestion. With over 8,000 active satellites and millions of debris fragments in Earth’s orbit, collision risks threaten global communication, navigation, and scientific missions. Traditional tracking systems often rely on expensive radar or optical solutions, limiting accessibility for operators like MDA Space Ltd. and emerging space agencies. The STARS system addresses this gap by introducing a passive phased array prototype that combines affordability with advanced real-time tracking capabilities, prioritizing scalability for future large-scale deployments.

At its core, the STARS system employs digital beamforming to steer a 2×2 phased array electronically, eliminating moving components and enabling rapid directional adjustments. This innovation, paired with FPGA-based signal processing, direction-of-arrival estimation, and Doppler shift frequency analysis, allows the system to track satellites precisely. The design integrates modular software subsystems, ensuring seamless scalability from a lab-tested prototype to a 20×20 array modeled in MATLAB.

As satellite deployments surge, the STARS system lays a critical foundation for democratizing advanced space situational awareness. It empowers operators to safeguard assets and sustain the long-term viability of Earth’s orbital environment.

Andrey Abushakhmanov ar.abushakhmanov@gmail.com; Anshul Israni anshul.israni01@gmail.com; Diya Thakkar diya25002@gmail.com; Kieran Ross kieranross5451@gmail.com; Nico Zhu nzhu2000@gmail.com

Project Client: Delta Controls

Project Description: Every year, undetected falls contribute significantly to injuries and hospital admissions, particularly among older adults and individuals with mobility challenges. Traditional monitoring methods, such as periodic checks, camera surveillance, or wearable devices—which can fail when a resident is unconscious—often fall short in providing timely intervention.

Our solution employs a 60 GHz Doppler Effect radar paired with a trained convolutional neural network to accurately detect falls. To ensure the system only activates in appropriate scenarios, a thermal sensor continuously counts occupants in the room. When it determines that only one person is present, the fall detection system is triggered, ensuring that irrelevant motion or multi-person activities do not lead to false alarms.

Our product sends an immediate notification to a dedicated screen that displays real-time occupancy data along with a clear alert for any fall incidents. This rapid, non-invasive approach allows caregivers and facility managers to monitor multiple rooms efficiently, ensuring swift response times and enhancing overall safety.

Project Description: Our client, Stantec, is a global leader in sustainable engineering, architecture, and environmental consulting. They offer design and engineering solutions in various sectors, such as buildings, water, energy, and transportation. 

In the buildings sector, a challenge faced by Stantec is the issue of harmonics generated by modern building electrical systems. These harmonics, caused by non-linear loads such as variable frequency drives (VFDs), can lead to overheating, equipment degradation, and energy loss in critical equipment like transformers and motors. Currently, electrical designers use harmonic filter protection in building electrical systems to mitigate the effects of harmonics. However, while the harmful effects of harmonic distortion have been studied, there is no defined threshold for the number of non-linear loads (i.e. VFDs) generating harmonics that damage other building loads such as LEDs. With a stronger understanding of VFDs and their harmonic distortion effects, Stantec will be able to optimize building designs, reduce operational risks, and address energy efficiency issues. Therefore, through this capstone project, we were able to determine how many VFDs without harmonic filter protection would cause harmonic distortion severe enough to damage external loads on the same electrical panel.

Prayetnaa Kansakar prayetnaa@gmail.com; Sara Han sarahan716@gmail.com; Khush Brar khushbrar01@gmail.com; Tommy Yoo tommyyoo123@gmail.com; Chenshun Hong contact@chenshun.me

Project Description: A critical challenge faced by the Remote Handling team at TRIUMF is ensuring the reliability and longevity of their commercial off-the-shelf cameras, which experience radiation-induced degradation over time. This leads to:

• Frequent camera failures
• Increased maintenance and replacement costs
• Potential safety risks, requiring personnel to interact with the damaged electronics

To address this, the project aims to develop a standardized, cost-effective methodology to evaluate the radiation resistance of five selected camera models under controlled exposure conditions. Testing will take place in an active area with a known radiation field, which represents the type of environment the cameras will be exposed to.

The result of the test will inform camera procurement strategies, ensuring future camera selections align with TRIUMF’s operational needs and radiation tolerance requirements. Additionally, the project will contribute to the definition of failure criteria, development of data acquisition tools, and integration of predictive modeling techniques to improve long-term camera reliability assessments.

Tiffany Zhuang tiffanyzhuang0515@gmail.com; Marcus Fung marcuscfung@gmail.com; Chengyu Yang k2066343831@gmail.com; Sei Ishikawa ishikawasei210@gmail.com; Lijin Gao gaolijin0807@gmail.com

Project Description: Our client, Creative Energy, is a district energy company who own and operate a steam plant in downtown Vancouver. This steam plant provides steam to a network of over 200 customers for heating purposes.

Our project is focused on providing our client with customer heating load predictions. These predictions can be used to inform the steam plant control system. Having access to load predictions will allow the client to optimize fuel consumption and improve the efficiency of their steam plant. 

The design problem was solved by creating a machine learning algorithm that could be trained on each customer’s specific heating profile to predict steam consumption based on weather forecasts.

Maddy Paulson https://www.linkedin.com/in/madelinepaulson/; Sara Hematy https://www.linkedin.com/in/sara-hematy-90625424b/; Youssef Mohammed https://www.linkedin.com/in/youssef-mokhtar1/; Brandon Just https://www.linkedin.com/in/brandon-just/; Aleksandar Stajic https://www.linkedin.com/in/aleksandarstajic/ 

Project Client: Alberni Yachts Inc.

Project Description: Diesel burning marine vessels are a large source of our world’s carbon emissions, with one gallon of marine diesel generating 21.24 pounds of CO2. Our team has been working with Alberni Yachts, a company aiming to reduce the world’s reliance on fossil fuels for ships by developing a fully electric, sustainable 65-foot luxury yacht capable of extended offshore use. 

Our team conducted a feasibility study on the electrical design of the vessel by sizing components, estimating loads and range, and simulating expected power generation under various test conditions. The renewable energy sources featured on the Alberni 65 include solar panels, wind turbines, and hydrogeneration to recharge the vessel’s energy storage system while it is offshore. Load profiles were created for different load scenarios based on factors such as: ambient temperature, wind speed, and daylight hours. Our team used a combination of solar irradiance data, MATLAB simulations, and AutoCAD shading analysis to estimate the power output of the solar panels. Similarly, for the wind turbines we used wind data and turbine power curves to estimate the power output. Additionally, different design options were analyzed for the vessel’s electrical distribution by comparing factors such as power efficiency, material and equipment costs, and system complexity.

Tessa Clement clementtessa@gmail.com; Jiahao Yan yanjiahao156@gmail.com; Yichi Zhang yichi0925@gmail.com; Ruihan Zhu ruihan03@student.ubc.ca; John Van der Star johnvandstar@gmail.com 

Project Description: Substations play a vital role in delivering electricity to communities, and as power demand increases, additional infrastructure—such as air-core reactors—is needed to maintain system stability. At BC Hydro, these reactors are installed to limit fault currents and overvoltages, enhancing the resiliency of the distribution system and making electrical equipment more cost-effective by reducing the electrical stresses they must withstand.

One challenge with air-core reactors is the strong magnetic fields they generate, which may impact sensitive electronic medical devices and other wearable technology used by workers. Measuring these fields at every substation is not practical due to the time and costs involved.

To address this, our project is developing an Excel-based magnetic field calculation tool that allows engineers to estimate the intensity of magnetic fields around air-core reactors quickly and accurately. By entering key parameters such as reactor size, current load, and distance, users can generate real-time calculations and visualize how the field strength changes with distance. The tool also accounts for shielding effects from surrounding materials, improving precision.

This tool will help BC Hydro plan substation expansions while ensuring compliance with safety standards by being the first step towards understanding the impact of magnetic field created by these reactors. 

Project Description: During a cardiac arrest, every second counts. Studies show that for every minute CPR is delayed, the chances of survival decrease by 7-10%. One of the critical challenges during CPR is accurately checking for a pulse, as manual pulse checks can be difficult and unreliable, especially in high-stress situations. Pausing chest compressions to manually check a pulse only increases the time without circulation, significantly raising the risk of mortality. Our project addresses this issue with the design of an innovative real-time pulse-detection patch—a non-invasive, portable, and rechargeable device that accurately discriminates between a live and non-live pulse.

The device is placed on the neck, where pulse detection is most sensitive, providing real-time feedback on pulse activity and heart rate with an accuracy superior to traditional manual palpation. Unlike many common devices, which become less effective during cardiac arrest due to peripheral hypoperfusion, our patch is specifically designed to maintain high sensitivity even under these conditions. This ensures that medical professionals can quickly assess the patient’s status without interrupting CPR, leading to better decision-making in critical moments.

Our solution incorporates advanced, medically validated signal processing techniques, while leveraging photoplethysmography technology. The device is designed to be both portable and reliable, making it an ideal tool for use in emergency medical situations. It also offers clear, immediate feedback via visual signals, ensuring that responders can easily interpret the data. With this device, we aim to reduce the time spent on pulse checks and improve the overall effectiveness of CPR, ultimately enhancing patient survival rates during cardiac arrest.

For any inquiries or potential collaborations, please feel free to contact the team at: travisinnes1600@gmail.com

Project Description: Our project demonstrates how a mobile banking app can leverage Azure Communication Services (ACS) to enhance client interactions for financial institutions. By integrating seamless calling and AI-powered conversation summaries, the app improves communication between clients and financial advisors. This ensures advisors have real-time client insights, streamlining interactions and enhancing the overall banking experience.

Major Design Contribution: A key technical challenge we tackled was integrating ACS with Copilot AI to generate real-time conversation summaries and client insights. This required optimizing API interactions, handling data privacy concerns, and ensuring smooth synchronization between voice calls and AI-generated summaries—all while maintaining a user-friendly experience in a mobile banking interface.

Contact Information: ilyve@student.ubc.ca

Project Description: The AI-Enhanced Security Testing Platform is designed to make security testing for web applications easy and accessible to everyone. It features a user-friendly web interface that simplifies the entire security testing process. This platform automates the execution of Dynamic Application Security Testing (DAST) tools, allowing users to quickly identify potential vulnerabilities.

Once the testing is complete, the platform generates a detailed summary report enhanced by artificial intelligence. This report highlights security issues while providing practical tips for implementation and answers any questions users may have about the findings. By streamlining the security testing process, this platform makes web application security more efficient and approachable for developers, regardless of their technical expertise.

Alfredo Alexei del Rayo alexeidelrayo@gmail.com; Andrew Piemonte andrewpiemonte@gmail.com; Junsu An anjjunsu@gmail.com; Ranbir Sarma ranbirs963@gmail.com

Project Client: Dependable Systems Lab @ UBC ECE

Project Description: Our client, Dependable Systems Lab, aims to develop Med-AI Sentinel, a new product to address the challenge of analyzing and securing machine learning-based medical devices. 

Medical device manufacturers face significant challenges ensuring the security of their cutting-edge machine learning enabled devices. The project aims to help medical manufacturing companies assess device vulnerabilities, investigate peripheral attack paths, and mitigate risks associated with their devices. The project scope includes a full-stack website that both end-users and admins utilize. The application creates detailed vulnerability reports that identify potential risks, provide relevant metrics, prioritize vulnerabilities, and deliver security-related information regarding the medical device.

Sally An sallyyan@student.ubc.ca; Sylvia Chen sylvia2002chen@gmail.com; Elisa Pan elisapan622@gmail.com; Vaghul Sivakumar vaghul12345@gmail.com; Vishrut Ohri vishrut.ohri@gmail.com

Project Description: Video streaming accounts for around 80% of internet traffic, making efficient video compression essential for everything from Netflix to Zoom calls. For NETINT Technologies, our client and a specialized provider of video encoding products, we improved video compression (encoding) for the AV1 format, a popular video format on the web. We created a deep neural network and added it to the AV1 reference software encoder. This network was trained on a computationally intense algorithm that fine-tunes how videos are compressed while maintaining quality: Rate-Distortion Optimization Quantization (RDOQ). 

Our neural network was carefully trained to perform the same adjustments as the algorithmic RDOQ process. Typically, RDOQ requires significant processing power. In addition to providing a faster hardware approach to algorithmic RDOQ, a neural network more efficiently utilizes hardware resources, decreasing power usage of AV1 video encoding hardware while maintaining video quality. We measured video quality using industry-standard quality measurements of our modified encoder to validate that our compression algorithm produces the expected output. This improvement could help deliver higher-quality video using less bandwidth.

Throughout this project, our team had to overcome several significant technical challenges. Most importantly, we had to develop a training methodology to create and tune a neural network that fits various video samples to ensure that video quality is maintained or improved. This required extensive training to ensure our project works across all sorts of content, from news broadcasts to action movies.

Nazare Infante Neal neneal@student.ubc.ca; Aaditya Suri asuri02@student.ubc.ca; Vincent Chernin vchernin@student.ubc.ca; Yash Parikh yashpar1@student.ubc.ca

Project Description: In collaboration with TELUS, our team developed the Emergency AI Response System (EARS) to address the critical issue of unanswered 9-1-1 calls during large-scale emergencies and natural disasters. When emergency call centers become overwhelmed, EARS utilizes machine learning to analyze and prioritize calls in near real-time, ensuring that the most urgent cases receive immediate attention.

Our system integrates signal feature extraction and speech-to-text technologies to assess both linguistic and acoustic features, determining the caller’s stress level and identifying urgent contexts. We designed a custom neural network that classifies 9-1-1 calls into high and low priority based on extracted audio and text features.

To demonstrate our model’s effectiveness, we developed a web-based proof of concept, showcasing its ability to classify emergency calls in near real-time.

Leo Kamino  leo.kamino@gmail.com; Tayyib Chohan tayyibchohan@gmail.com; Deepan Chakravarthy kdeepan240@gmail.com; Fei Kuan kuan@iatfei.com; Mabel Wang mabelwang20020410@gmail.com

Project Description: Our team is working for the UBC Manufacturing Engineering department to create Augmented Reality (AR) virtual learning tools for students. The Interactive AR Factory app allows students to see and interact with virtual machinery and production lines, allowing them to gain experience in manufacturing environments without the need of physical machinery. Professors and TAs will be able to customize and modify these virtual environments to suit student’s specific learning goals and enhance the breadth of their lab experiences.

This year, we integrated a physical Programmable Logic Controller (PLC), which allows for physical inputs to be registered in the virtual world. Integrating virtual objects with real-time environments in a multiplayer system was also one of our major achievements. Also, we implemented the interactions between physical hands and virtual objects, which greatly enhance the immersive lab experience.

Project Description: In Metro Vancouver, where 88 million pounds of textiles are discarded annually, the gap between clothing waste and local processing capacity has created an urgent need for sustainable solutions. Sorty, developed by our team for Dreamstill Technologies, addresses this challenge by empowering users to make informed decisions about their unwanted clothing. Designed for sustainable fashion enthusiasts and environmentally conscious individuals, Sorty harnesses a decision tree algorithm to assess the condition of clothing items accurately, provides clear recommendations — resell, donate, repair, or recycle — and connects users to a curated map of local services including consignment stores, thrift shops, donation centers, and repair facilities.

A core technical challenge was designing an accurate clothing condition assessment system despite limited datasets of damaged or worn textiles. To overcome this, we embedded a decision tree algorithm into the app to guide the user through the assessment of the clothing condition. At the same time, we conducted research and explored the potential of machine learning models for assessing the clothing condition. We delivered a machine learning research report that offers an in-depth investigation of current clothing datasets, a comparison among existing applicable models, feasibility analysis on alternative models and APIs, and detailed cost estimations for training or implementing these models. This resource not only supports the current implementation but also serves as a strategic blueprint for our client’s future data collection and model enhancement efforts.

For further information or to discuss potential collaborations, please reach out at pscholtens2001@gmail.com, hardyh074@gmail.com, yebinjie528@gmail.com, amoghkumar2301@gmail.com.

Project Description: The SEC mandates that all publicly traded companies submit filings on a regular basis. These filings contain valuable financial information; however, they are lengthy and are not structured in a standardized way. Thus, obtaining insights from these filings is a time consuming and complex task. To resolve this, our team utilized LLMs and parsing to extract important insights from these filings in a structured format, reducing the manual workload required to analyze these documents.

Project Description: Forests are the world’s largest terrestrial carbon sink, and the demand for accurate monitoring is growing. Automating forest surveys with drone technology improves efficiency, but flight planning often requires navigating multiple complex tools that lack customization for organizations conducting these surveys.

Our project simplifies forest flight planning with a single application with an intuitive interface. Users can generate flight paths by drawing a polygon, and our algorithm automatically creates an efficient back-and-forth route covering the area. Customization options include altitude adjustments, path rotation, and selection of takeoff points and observer locations. The planned route can be visualized in a 3D map, saved for future access, and exported to FlyLitchi’s drone flight software.

Project client: Holdr

Project description: To begin with, Holdr is an online fan club platform where music artists can connect and receive direct support from their fans. With Holdr, followers can subscribe to an artist’s Holdr Club, granting them access to the artist’s exclusive content. However, once an artist’s career ends, these memberships lose their value, leaving fans with no ongoing benefits. This lack of long-term value for passionate supporters from becoming members of an artist’s Holdr Club, limiting the platform’s ability to retain users and generate sustained revenue.

Our team PB-27’s project, the Artist Time Vault, will introduce a new feature on the Holdr platform that aims to provide lasting value for Holdr Club members. Artists can upload files to the Artist Time Vault, where they are securely stored in an external database and remain hidden from fans initially. Over time, the artist can choose to release selected contents, making them visible to members of the artist’s Holdr Club, even after retirements. This feature not only helps to sustain fan engagement but also opens new opportunities for monetization through the membership trades, ensuring benefits for both the artists and their supporters in the long run.

Our primary technical challenge revolves around security, database design, and logic handling for release contents. Firstly, since the files and folders need to be securely stored before release, robust user authorization and authentication are necessary, along with support for 2 access levels: fan and artist.. Secondly, the database must accommodate a directory-like folder structure (i.e, nested folders) with minimal redundancy. Lastly, releasing a file or folder can trigger cascade releases, requiring careful handling. For example, since it is natural for users to be able to see the parent folder of a released content, releasing a file will cascade release all of its parent folders. This introduces complexities for edge cases such as moving a released file into an unreleased folder, and also access permissions check.

Tony Shi tonyshi12@gmail.com; Xuan Tung Luu xuantung.brian@gmail.com; Yibo Chen yibochen2019@gmail.com; Muhan Li muhanli@student.ubc.ca; Divy Patel divy07ubc@gmail.com

Project Description: The Tree Species Identification App utilizes advanced machine learning techniques to automate tree identification from remotely sensed data. This project addresses significant challenges in forestry management, ecological monitoring, and conservation, particularly the high costs and time required for manual tree identification in remote or inaccessible areas. Our solution supports conservationists, researchers, and forestry professionals by providing a rapid, accurate, and cost-effective approach to biodiversity assessment and forest resource management.

Major Design Contribution: The main technical challenge was integrating diverse data types—including mobile imagery, drone-based RGB imagery, and LiDAR point clouds—into a unified identification process. An additional hurdle was developing effective machine learning models despite having limited prior expertise in forestry and tree identification, along with a scarcity of labeled datasets. To overcome these challenges, we developed specialized models capable of highly accurate classification of tree species and genera.

Key Components:

• Ground-based Segmentation (YOLOv8): Segments individual trees from mobile-captured images.
• Ground-based Classification (ResNet): Classifies tree species from segmented images captured via mobile devices.
• Drone-based RGB Classification (ResNet): Classifies tree species using RGB imagery from drones.
• Drone-based Tree Segmentation: Identifies and segments trees in drone imagery for detailed analysis.
• Drone-based LiDAR Classification (DGCNN & Metadata Fusion): Accurately classifies tree species using detailed LiDAR data and metadata fusion.

Each model was independently trained and thoroughly validated, ensuring robustness and reliability across various environmental conditions and data collection methods.