Doctoral Researcher, Artificial Intelligence in Medical Imaging / Signal Analysis Lab
Trustworthy AI systems for evidence-heavy work.
A technical portfolio for collaborators, hiring teams, and organizers who need to inspect the research agenda, system builds, talks, and writing behind my work in AI agents, speech intelligence, cybersecurity, and regulated deployment.
Research base
Doctoral research in the Artificial Intelligence in Medical Imaging / Signal Analysis Lab at NYCU.
Speaking signal
Official CYBERSEC 2026 breakout session on cybersecurity design for AI software medical devices.
Publication activity
Two English CISC 2025 conference papers connecting security analysis with practical deployment and defense.
Triton makes the next step obvious for developers. This site should do the same for collaborators, reviewers, hiring teams, and event organizers.
The redesign borrows Triton's clarity: a direct technical promise, a compact capability list, getting-started paths, and evidence sections that let visitors inspect what has been built, written, and presented.
Evidence enters through speech, text, threat, and research sources, then moves through grounded AI workflows before a human review or deployment decision.
audio, transcripts, evidence, threat signals
retrieval, agents, extraction, system evaluation
traceability, uncertainty, human approval
security, governance, regulated workflows
Agent workflows should expose what they used, what changed, what remains uncertain, and where a human should intervene.
Long-form audio needs more than transcription: timestamps, speaker turns, retrieval, extraction, and reviewable outputs must stay connected.
High-stakes AI work needs threat models, privacy assumptions, audit trails, rollback paths, and governance from the beginning.
The homepage now treats every artifact as part of the same library: case studies show implementation evidence, talks show public technical framing, and writing records the reasoning that connects them.
The work asks how AI agents and decision-support systems stay reliable, reviewable, and useful under real operational constraints.
Designing AI agents and systems where reliability, evaluation, human review, and traceability are part of the architecture rather than afterthoughts.
Building ASR + LLM + RAG pipelines and agent workflows for conversational analysis, structured extraction, and evidence-aware reasoning over long-form audio and transcripts.
Studying privacy, leakage, adversarial risk, and governance constraints that shape AI systems used in regulated or security-sensitive environments.
Each project explains the problem, the system design, and the evidence that matters in practice.
A research case study on federated learning privacy leakage, gradient inversion risk, and defense trade-offs for sensitive collaborative training.
Readiness
Research seedA security research direction with public threat-model framing and supporting references.
Last reviewed: May 7, 2026
Doctoral Researcher
Open case study
A research-led case study on retrieval-augmented fraud conversation analysis, designed to keep LLM outputs grounded in transcript evidence for high-stakes review.
Readiness
Research seedA research-led case study framing grounded RAG as an analyst review workflow.
Last reviewed: May 7, 2026
Doctoral Researcher
Open case study
An evidence-aware speech intelligence pipeline using ASR, retrieval, and LLM extraction to turn long-form conversational audio into structured, reviewable outputs.
Readiness
PrototypeA public case study for an evidence-aware ASR, retrieval, and LLM extraction workflow.
Last reviewed: May 7, 2026
Doctoral Researcher
Open case studyThe labels separate early research, concrete prototypes, paper-backed work, production-style case studies, and archived artifacts so visitors know how to read each project.
A scoped research direction with a clear question, references, and next evaluation work.
A system or pipeline direction concrete enough to discuss as an implementation path.
A project connected to a publication, formal submission, or conference-facing artifact.
A polished system narrative with deployment assumptions, user workflow, and risk controls.
A finished or historical artifact kept for reference rather than active development.
Short essays and research notes make the design choices behind the technical work easier to inspect.
Liquid Glass, visionOS, and AI point toward a future where the operating system is redefined less by the kernel than by a new operating surface for search, orchestration, and cross-app work.
A research-oriented essay on cross-node fraud intelligence, minimal disclosure, and trustworthy AI design for high-stakes pattern formation under fragmented evidence.
A research note on building speech evidence pipelines with ASR, retrieval, and LLMs while keeping outputs grounded in transcript evidence.
Speaking spans official sessions and English conference papers across AI agents, cybersecurity, speech systems, and deployment work where reliability and traceability matter.
Breakout session on cybersecurity practice for AI software medical devices, using FDA 524B to connect threat modeling, SBOM, Zero Trust design, and auditable risk governance in heavily regulated environments.
English conference paper examining how AI-era RaaS operations evolve through BYOVD, LOTL, covert C2, and adaptive tradecraft, then mapping those threats to a ZTAID-grounded zero-trust defense strategy.
English conference paper proposing Threat Pulse Modeling (TPM) as a way to translate live cyber threat intelligence into ZTAID maturity signals for continuous zero-trust assessment.
How to think about reliability, evidence, human review, and system behavior when AI is used in environments where mistakes carry real cost.
Research groups, labs, interdisciplinary audiences
How modern RaaS campaigns combine automation, BYOVD, LOTL, and covert C2 techniques, and how ZTAID-aligned zero-trust strategy can structure practical detection, containment, and recovery.
Cybersecurity conferences, blue teams, graduate seminars
How cyber threat intelligence can be translated into pulse events, ZTAID pillar scores, and measurable maturity signals to support faster defensive adaptation.
Cybersecurity researchers, zero-trust programs, graduate seminars
Design patterns for speech and language pipelines that move from raw transcripts to grounded, inspectable outputs in analyst-facing settings.
NLP teams, speech researchers, applied AI practitioners
How to design tool-using agents and coding-assistant workflows that stay inspectable, grounded, and useful for real work.
AI product teams, research labs, engineering groups
Why privacy, leakage risk, adversarial thinking, and deployment assumptions should be treated as core system questions rather than compliance afterthoughts.
Security teams, engineering groups, policy-adjacent stakeholders
The work combines doctoral research, agent building, and an investigation-informed view of evidence and failure.
I bring together medical AI lab research, system building, and investigation-informed thinking to design AI agents and IDE-like agent systems that stay useful when evidence, regulation, and deployment constraints matter.
Role
Doctoral researcher at NYCU
Focus
AI agents, speech intelligence, and cybersecurity
Approach
Evidence-aware agents and systems for real-world use
I welcome thoughtful conversations around AI agents, IDE-like agent systems, trustworthy AI, speech and language systems, and deployment in security-sensitive or regulated environments.
Email is the fastest route. The contact page also shows the context that helps most.
Research collaboration and interdisciplinary lab conversations
Speaking invitations for AI agents, speech systems, or cybersecurity
Hiring, technical peer exchange, and IDE-like agent system design
Helpful first message
Share your context, why the conversation fits, and what kind of next step would be most useful.