Research evidence
Turning manuscript and benchmark work into claim-gated evidence surfaces: CDS-ASR, PB-EGP/STV, TFSC false governability, and WISA CaseTrace.
Now / Recent work
April-June 2026
I build and study AI systems where the hard part is not only model output, but evidence, review, governance, and safe handoff: speech decision stability, clinical workflow support, agent governance, cybersecurity evidence, and practical AI systems education.
Current focus
Deployable prototypes
Building narrow, reviewable systems that demonstrate the workflow before making larger claims: UroPrevisit Navigator, AI Triage Kiosk, Jarvis Voice Sight, and Project AURA.
Teaching and translation
Packaging AI systems engineering into practical learning artifacts: AI Gateway, agent governance, model serving, voice AI, cybersecurity, and enterprise delivery.
Recent highlights
Teaching
AI Systems Engineering Handbook
Built a tutorial-first handbook around enterprise AI systems: foundations, Linux/cloud/K8s/GPU, LLM apps, RAG, AI Gateway, agent governance, voice AI, security, delivery, and AI-assisted engineering.
13 modules, accelerator packages, teacher packets, and validation checksResearch
CDS-ASR / JANUS evidence hardening
Advanced a speech-to-decision stability research line that tests whether downstream decisions remain stable under plausible ASR alternatives.
Selected-300 aggregate review complete; raw audio and transcripts remain privateResearch
PB-EGP and Support-Transfer Validation
Developed provenance-bounded evidence graph packet work for small-model decision support, with release gates, audit protocols, and public-benchmark orientation.
Fallback package and validation gates ready for owner release-channel decisionSystem
AI Triage Kiosk demo
Built a synthetic vital-aware intake and staff-review summary demo for a June market demonstration, with a narrow API contract and explicit clinical-scope controls.
Synthetic-data market demo; no autonomous diagnosis or production clinical triageSystem
UroPrevisit Navigator
Hardened a urology previsit workflow prototype for adaptive question navigation, missing-field repair, clinician-review summaries, and PSA/CRM-ready proposal framing.
Synthetic-data workflow prototype for staff-review supportSystem
Jarvis Voice Sight
Built a mock-first realtime voice agent path with always-listening interaction, VAD, barge-in, turn isolation, streaming TTS, and Ollama/vLLM runtime support.
Realtime prototype with typecheck, lint, test, smoke, health, and demo gatesSpeaking
CYBERSEC 2026 AI SaMD cybersecurity talk
Delivered a 30-minute session on cybersecurity practice for AI software medical devices, connecting FDA 524B, threat modeling, SBOM, Zero Trust, and auditable risk governance.
Delivered May 6, 2026Research
TFSC false-governability manuscript
Rebuilt an AI governance manuscript around false-governability onset, review scarcity, bounded authority, and reconstructable runtime governance.
Submission-candidate package with public-safe evidence framingMonth-by-month
April 2026
Research and public-profile foundations: journal-path revision, urology demo readiness, governed scam-content research routing, Network Security and coursework packaging, and CYBERSEC talk preparation.
- Prepared evidence-led public-site and resume direction around trustworthy AI and speech systems.
- Moved urology previsit work toward a safe local demo and reviewer-ready workflow evidence.
- Kept Meta/Threads research bounded as reviewer-support and public-interest scam-risk work.
May 2026
Clinical workflow prototypes and manuscript reframing became the center of gravity: AI triage, UroPrevisit Navigator, Project AURA, TFSC, WISA, and CYBERSEC delivery.
- Delivered CYBERSEC 2026 medical cybersecurity session.
- Converted imedtac AI triage into a narrow synthetic vital-aware demo with a staff-review summary boundary.
- Clarified the Health Taiwan / urology proposal route as PSA follow-up, adaptive previsit questions, clinician-review summaries, and CRM-ready fields.
June 2026
The work shifted into evidence hardening, release gates, and education packaging across CDS-ASR, PB-EGP/STV, TFSC, WISA, Jarvis Voice Sight, Project AURA, and the AI Systems Engineering Handbook.
- Completed aggregate-safe research evidence surfaces for speech decision stability and evidence-packet work.
- Built AI systems engineering teaching packets around AI Gateway, agent governance, model serving, vLLM/SGLang, and enterprise delivery.
- Advanced realtime voice interaction and local meeting-summary trust prototypes with explicit runtime validation gates.
Public boundary
Recent work is summarized from public-safe repo artifacts and planning summaries. Raw planning notes, private contacts, patient-like data, raw transcripts, credentials, and patent-sensitive mechanics are intentionally kept out of the website.