Research
Evidence-aware AI systems for high-stakes work.
My research asks how AI systems preserve evidence, human review,
reconstructability, and public-safe scope when they move from fluent
outputs into decisions, workflows, and action.
CDS-ASR Manuscript hardening
A speech-to-decision stability research line that asks whether downstream decisions remain stable under plausible transcript alternatives.
ASR quality should be evaluated not only by word error, but by whether plausible recognition alternatives change downstream decisions in high-stakes conversational workflows.
JANUS workspace organized around audio, ASR hypotheses, risk atoms, counterfactual variants, CEIS scoring, and constrained recovery.Selected-300 aggregate review surface completed with dual-review and model-level assessment summaries.
PB-EGP / STV Benchmark design
Research on evidence graph packets that help small models make more stable, better-grounded decisions under the same token budget.
For sub-10B models, the decisive question is often not more context, but better evidence discipline: source-bounded packets, valid provenance, and stable decision transfer.
PB-EGP scope defines provenance-bounded evidence graph packets against top-k text, summaries, graph pruning, and context compression.Support-Transfer Validation work advanced through adjudication, fallback packaging, release manifests, and semantic validation.
False Governability Submission candidate
A technology-governance research line on how action-capable AI can appear governable while reconstructability collapses underneath.
AI action scaling creates a review-scalability gap: organizations may preserve visible governance signals while losing the ability to reconstruct authority, evidence, review, and accountability.
TFSC-facing manuscript rebuilt around false-governability onset under review scarcity.Taiwan 165 is used as a public-safe parameterization anchor for high-audit governance pressure.
CaseTrace Conference paper path
A security-evidence research direction that turns reviewer feedback into traceable public-source case evidence, uncertainty labels, and baseline comparison.
Security papers become stronger when claims are mapped to reproducible case evidence, uncertainty, defense-control implications, and explicit comparison instead of broad threat narratives.
WISA 2026 package reframed around a new contribution-first CaseTrace direction rather than a polished resubmission.Evidence package records 18 coded rows across 3 cases, source corpus, case index, missing-evidence notes, baseline comparison, and LNCS structure.