Michael Nakagawa
Research Algorithm & AI Engineer Philips
A Research, Algorithm, and AI Engineer at Philips Healthcare, he brings over two decades of experience spanning emergency care, patient monitoring, and innovation strategy. He specializes in ECG analysis, AI-driven algorithm development, software systems architecture, FDA submissions, and verification and validation, advancing clinically robust, regulatory-compliant healthcare technologies.
Seminars
Tuesday 21st July 2026
Workshop C: Systems-thinking Simulation Exercise: How Much Autonomy Is Too Much? Designing Agentic AI for Regulated Medical Software
2:00 pm
A collaborative simulation where participants design and test autonomous AI agents, exploring how goals, constraints, and decision making logic shape system behavior.
- Teams confront a realistic scenario where an autonomous AI agent can write code, review PRs, run tests, and respond to cyber threats, forcing them to decide what actions they’re truly comfortable delegating
- Participants break into governance style groups (engineering, V&V, regulatory, cybersecurity, and patient safety) to define guardrails around autonomy, validation, documentation, risk, and oversight across the SDLC
- Live “curveballs” expose failure modes and challenge assumptions, culminating in a cross-discipline debate on what to allow, forbid, or reconsider, revealing blind spots and making agentic AI risks tangible
- Decide whether autonomous AI agents are advantageous in comparison to classic AI/ML models (such as Co-pilot)
Closing Question: In 5 years, which of today’s ‘absolutely not’ decisions will sound overly cautious?
Wednesday 22nd July 2026
Panel: Proving AI Safety Through Robust Validation Beyond Performance Metrics
12:15 pm
- Understand why AUC, sensitivity, and specificity are insufficient to demonstrate clinical safety
- The “no gold standard” problem: disagreement among clinicians as a validation challenge
- Identifying bias, edge cases, and underrepresented populations in training/testing data
- Integrating AI validation with risk management, V&V, and regulatory evidence expectations
Wednesday 22nd July 2026
Managing AI Model Evolution While Maintaining Regulatory Control
9:45 am
- Recognize risks introduced by model drift, data drift, and real-world performance degradation
- Identify and manage governance challenges with rapidly iterating AI versions and generative tools
- Learn monitoring strategies for “unlocked” or adaptive models in the post-market phase
- What performance, safety, and usage signals regulators increasingly expect teams to track
