About me
Hi!, I am a PhD researcher at the University of Oulu (Faculty of Information Technology and Electrical Engineering), supervised by Dr. Nirnaya Tripathi, Prof. Minna Isomursu
and Dr. Rahul Mohanani, funded through the FAST Finnish Software Engineering Doctoral Research Network.
My work sits at the intersection of software engineering, AI governance, and clinical decision support. I study how self-adaptive AI systems can be built to be safe, accountable, and practically deployable in clinical settings, while still being flexible enough to be useful. The core problem I keep coming back to is that AI systems which adapt at runtime can genuinely improve care, but they also introduce governance risks that existing software safety techniques were not built to handle. My thesis tries to develop the engineering foundations needed to close that gap, covering architecture design, constraint mechanisms, and evaluation under realistic clinical conditions.
Outside the thesis work, I am broadly interested in runtime AI governance, privacy and security in adaptive health systems, and the socio-technical side of getting AI into nurse-led and clinician-led workflows in practice.
Let’s Connect!
Revised Research Title
Self-Adaptive AI Systems for Clinical Decision Support: Towards Safe, Governable, and Privacy-Preserving Runtime Adaptation
Research Questions
RQ1 What are the existing approaches, engineering gaps, and governance challenges in self-adaptive AI systems for clinical decision support? (Addressed in Paper I)
RQ2 How can software engineering constraint mechanisms be applied to an adaptive AI architecture to enforce safety, data integrity, and auditability in clinical decision support? (Addressed in Paper II)
RQ3 How does a constraint-based adaptive CDSS architecture hold up under realistic clinical stress conditions, such as data drift, adversarial feedback, and varying load? (Addressed in Paper III, planned)
RQ4 How do clinicians perceive, trust, and interact with a governable adaptive CDSS prototype, and what shapes whether they rely on its recommendations appropriately? (Addressed in Paper IV, planned)
Research Objectives
Objective 1: Establish the state of knowledge (Paper I). Systematically review the literature on self-adaptive AI in medical software to map out where the engineering and governance gaps actually are. This gives the rest of the thesis a grounded starting point rather than an assumed motivation.
Objective 2: Design a safety-constrained adaptive architecture (Paper II). Using Design Science Research, develop and evaluate an architectural artifact that extends the MAPE-K adaptation loop with a Constraint Layer. The layer enforces four mechanisms: parameter bounds (C1), validation gates (C2), evidence anchoring (C3), and audit/rollback (C4). The design is validated through formative evaluation with clinical and technical stakeholders.
Objective 3: Prototype and stress-test the architecture (Paper III). Build a working prototype and evaluate it through simulation, focusing on how the system behaves under data drift and adversarial feedback. An ablation study will isolate the contributions of each of the four constraint mechanisms relative to a fully unconstrained baseline.
Objective 4: Study how clinicians interact with the system (Paper IV). Run a mixed-methods study with clinicians using the prototype in context, looking at behavioral reliance, trust calibration, and usability. Thematic analysis will add qualitative depth to understand how governance transparency affects clinical perception of the system.
Research Methodology
Recent Projects
MentalEEG A tool for EEG-based mental workload detection, developed as part of early exploratory work on AI-driven health monitoring.
SurveyFlow A platform for structured survey data collection, designed for use in research and clinical data-gathering contexts.
Prescription Management System (Ray PMS) A full-stack system for managing prescription workflows, linking patient records with medication tracking.
FamilySpend A personal finance tracking application, built as part of broader full-stack software engineering practice.
FitStreak A subscription-based fitness app built around the idea that fitness software should adapt to your life, not the other way around. It generates personalized workouts based on your goals, available time, and equipment, syncs automatically with Google Health Connect/Apple Health, and uses a streak system designed to keep you consistent without punishing you when life gets in the way. Built through 25 test releases and currently in closed beta on Google Play, with an iOS version in progress.
Publications
Accepted
Wendakoon, K. and Tripathi, N. Self-Adaptive AI in Clinical Decision Support: A Systematic Review. Nordic Conference on Digital Health and Wireless Solutions (Nordic Digihealth 2026), Oulu, Finland, June 2026. (Proceedings forthcoming, Poster 24)
Wendakoon, K. and Tripathi, N. A Safety-Constrained Architecture for Adaptive Clinical Decision Support. Nordic Conference on Digital Health and Wireless Solutions (Nordic Digihealth 2026), Oulu, Finland, June 2026. (Proceedings forthcoming, Poster 23)
In Progress
Paper III: Prototype and simulation evaluation of the constraint-based adaptive CDSS architecture. Target venue: Empirical Software Engineering (EMSE) or Journal of Systems and Software (JSS). Expected submission: late 2026.
Paper IV: Mixed-methods clinician study on trust calibration and reliance in a governable adaptive CDSS. Target venue: JAMIA or Journal of Biomedical Informatics (JBI). Expected submission: early 2027.
My Life in Oulu
