Posted 11:54 a.m. Thursday, June 4, 2026
Software engineering student develops AI models and ethical frameworks for safer healthcare systems
Artificial intelligence is becoming increasingly common in healthcare, helping providers with diagnoses, treatment decisions and patient care. In 2024, 71% of hospitals reported using predictive AI integrated with electronic health records, according to a study published in the National Library of Medicine.
But as healthcare systems rely more heavily on AI, UW-La Crosse Alumna Kaaviya Saraboji believes developers must focus on more than accuracy alone.
“AI should not just produce results,” Saraboji says. “It should be safe, fair, transparent and accountable.”
Saraboji, who graduated from UWL in May with a master’s degree in Software Engineering, is developing a lifecycle-aligned governance framework designed to help healthcare AI systems become more ethical and trustworthy from the very beginning of the development process.
Her research focuses on integrating ethical safeguards throughout the entire AI lifecycle — from planning and design to deployment and ongoing monitoring. Rather than evaluating ethics only after a system is already in use, her work aims to build those considerations directly into how healthcare AI is created.
Operationalizing ethical principles
Saraboji’s latest paper, titled “Operationalizing WHO Ethical Principles for Healthcare AI: A Lifecycle-Aligned Governance-by-Design Framework,” developed under the mentorship of Dr. Shivaram P. Arunachalam of Mayo Clinic and Dr. Dipankar Mitra, assistant professor of Computer Science & Computer Engineering, has been accepted for presentation at the Mayo Clinic AI Research Summit in Rochester, Minnesota, in June. The full paper is also under review for publication in the journal “AI in Medicine.”
Earlier in her graduate research, Saraboji developed the ARISTOTLE project — named after the ancient Greek philosopher — as a conceptual ethical readiness assessment tool for healthcare AI systems. The project explored ways to translate broad WHO ethical principles into measurable and actionable evaluation mechanisms for developers and healthcare systems.
The project is built around six ethical principles established by the World Health Organization (WHO) for AI in healthcare:
- Protect autonomy
- Promote well-being, safety and the public interest
- Ensure transparency, explainability and intelligibility
- Foster responsibility and accountability
- Ensure inclusiveness and equity
- Promote responsive and sustainable AI
While Saraboji found the WHO principles valuable, she says developers often struggle with how to apply them in practice.
“These ethical principles are high level,” she explains. “When it comes to practical implementation, there is a gap.”
The ARISTOTLE project generated ethical readiness assessments and recommendations intended to help organizations evaluate whether AI systems were being developed responsibly and safely. The project earned first place in the at-large WiSys Innovation On-Ramp Showcase in November 2025.
That work later helped shape Saraboji’s broader research on lifecycle-aligned governance frameworks for healthcare AI, focusing on embedding WHO ethical principles throughout the AI lifecycle — from planning and design to deployment and ongoing monitoring.
“The goal is to make ethical evaluation more consistent, transparent and easier to apply in healthcare systems,” she says.
Addressing real-world healthcare risks
Saraboji’s interest in healthcare AI ethics began during the summer of 2025 while working on a machine learning project focused on chronic kidney disease prediction.
At the time, she began asking broader questions about how the patient data had been collected, whether it was representative and unbiased, and how reliably the AI system would perform across different patient populations.
“I started asking questions about where the dataset came from, whether it was fair and unbiased, and whether the AI system would perform reliably across different healthcare settings,” she says.
She began studying reports and governance frameworks from organizations including the WHO, the European Public Health Alliance, the Food and Drug Administration (FDA), and the Organisation for Economic Co-Operation and Development (OECD), later completing training programs in healthcare AI ethics and HIPAA compliance, including the WHO’s Ethics and Governance of Artificial Intelligence for Health program.
The experience also revealed that AI systems that perform well during development may not always perform reliably in real-world clinical settings due to differences in patient populations, clinical practices and data quality. In some cases, healthcare AI systems may perform less accurately for underrepresented groups, provide predictions that clinicians cannot easily interpret, or become less reliable over time as healthcare environments and patient data change.
For example, healthcare systems in different countries or hospitals may use different clinical practices, lab values or patient populations. Without broader testing and monitoring, AI systems can become unreliable or inequitable when deployed in new settings.
Saraboji’s framework addresses these concerns by embedding governance mechanisms across the AI lifecycle, including data collection, model development, validation, deployment and post-deployment monitoring. The framework emphasizes external validation across different healthcare settings, fairness evaluation across patient groups, explainability for clinicians, and continuous monitoring to help ensure healthcare AI systems remain safe, reliable and trustworthy over time.
“If an AI system predicts a disease, clinicians should be able to understand how it reached that conclusion,” she says. “Transparency and explainability are important for building trust in healthcare AI.”
She emphasizes that healthcare AI should support — not replace — medical professionals.
“The goal is to help doctors make better-informed decisions and improve patient care,” she says. “AI should assist clinical decision-making while keeping human oversight at the center.”
Building better AI orchestration
Saraboji’s work extends beyond ethical assessment tools and governance frameworks.
In a separate project called AIOrch, she is exploring ways hospitals can better coordinate, monitor and manage the growing number of AI systems used across healthcare organizations. The project focuses on the lack of centralized visibility and oversight of healthcare AI systems — an issue that can create operational, clinical and orchestration challenges when multiple AI tools are used across departments.
AIOrch is designed as a centralized orchestration platform that tracks AI systems throughout their lifecycle, including development, deployment and monitoring. The project also explores how hospitals can improve auditability, workflow coordination and oversight of AI tools used in clinical environments. Presented at the 2026 WiSys Big Idea Tournament, The AIOrch project earned second place.
Both projects connect to her broader goal of building more responsible, transparent and well-governed AI systems for safety-critical environments.
“My research broadly focuses on applying machine learning to safety-critical systems, particularly in healthcare and road safety,” Saraboji says. “As the field continues to evolve, I became increasingly interested in ethical and responsible AI.”
Research with real-world impact
Saraboji says UWL’s software engineering program helped shape her research mindset by emphasizing practical, hands-on learning experiences.
“When I arrived at UWL, I was surprised by how practical and hands-on the assignments and projects were,” she says.
That experience led her to earlier research projects in wildlife and transportation safety, including machine learning approaches designed for wildlife road safety and animal-vehicle collision prevention.
One of her papers on AI-powered animal-vehicle collision prevention was published in "Electronics." Two additional papers focused on wildlife road safety and transportation safety analytics were accepted for presentation at the IEEE International Conference on Electro Information Technology (EIT). Another study on deer-vehicle collision prediction and resource-aware transportation safety planning using Wisconsin Department of Transportation data was submitted to the IEEE Global Humanitarian Technology Conference.
“Since I started this program, whenever I see a problem, I think about how technology and software systems could help solve it more efficiently,” she says.
Saraboji credits faculty mentors Mitra and Mao Zheng, professor of Computer Science & Computer Engineering, for encouraging her research and presentations throughout the process.
“Whenever I had questions, Dr. Mitra would always support me and say, ‘You should participate,’” she says. “He consistently encouraged me to participate in conferences, competitions, and presentations. Dr. Zheng has also been incredibly supportive and continues to check in on me and encourage me.”
Saraboji has been accepted into the doctoral program in computer science at the University of Idaho.
As healthcare systems continue integrating AI into patient care, Saraboji hopes her work will help ensure those technologies remain innovative, ethical, equitable and trustworthy.