Production AIthat ships.
Range and depth,
by deliberate design.
Clinical AI
Twenty-one years of clinical, education, and admissions work — anesthesia, psychiatry, special education. Domain depth that informs every system I design.
Agentic AI
Production agents on Claude's tool-use API and MCP. ReAct orchestration, multi-step planning, and evaluation harnesses with adversarial test sets.
Design-engineering
Working prototypes wired to real systems. Iterative bake-offs with Claude and Cursor, with the visual and interaction craft pushed until it ships.
Responsible AI
Fairness audits, bias detection, and SHAP interpretability. Programmatic labeling and weak supervision as production patterns — humans in the loop.
Flagship systems first.
Patent-pending clinical AI, a hackathon-shipped agent-governance platform, citation-grounded RAG, and DOI-published research — each solving a specific problem at the intersection of expert domain knowledge and machine intelligence.
Four questions every flagship system here is built to answer.
Who is allowed to change what, and who approved this change?
Compass-BlackBox IQ
What did it do, why, and can I reconstruct that later?
GUSS
How do I know it is right, and what is the pass condition?
Provenance
What are the parts, and are they swappable?
Governance Drift Researcher
Governance Drift Researcher
Agent governance · evidence-verified findings · CI gate
Live · PyPI + WeaveMind Cloud
Vendor model lifecycles move; the approved agent inventory doesn't. This scans an Azure AI Foundry catalog against a human-owned baseline and reports the drift under rules that make the report worth trusting: every finding carries its source URI, JSON field path, and content hash, and any finding whose citation no longer resolves is dropped and counted rather than published. Coverage gaps are stated, not implied. It exits nonzero on surviving drift so it works as a CI gate, and emits SARIF for Defender and code scanning. Also ported to Weft and run end-to-end on WeaveMind Cloud, where the human approval gate is structural — there is no API by which the agent can approve its own report.
Triton Kernel Lab
GPU kernels + compiler IR · study artifacts
Public · CI green on every kernel
Triton kernels — elementwise, softmax, matmul — written to be read as much as run. Each one commits its own lowering artifacts: TTIR, TTGIR, LLVM IR, and PTX, so the compiler's choices are inspectable rather than asserted. Benchmarks put the matmul against a hand-written CUDA C++ baseline and cuBLAS, with the correctness gate bounded by cuBLAS's own error against fp64 rather than an arbitrary tolerance, and softmax against torch.compile's steady state with a committed Dynamo explain artifact. The Jetson bring-up log records six failures in the order they happened, including the one an external cold read caught in the repo's own message.
GUSS
Edge inference appliance · headless + autonomous · solo build
Running · public case study
A governed autonomous agent on 7 watts. A hand-assembled, permanently headless Jetson Orin Nano serving a local Qwen3-8B at 12 tok/s that monitors its own health, restarts its own services, paper-trades a benchmarked portfolio inside code-enforced guardrails, and publishes its own dashboard — twice daily, unattended. The build's core lesson, earned across four documented model failures: code decides, the model narrates. Architecture, ADR-style decision log, and an honest build log in the public case study.
Compass-BlackBox IQ
Microsoft Foundry + MCP · solo build
Agents League · AI Skills Fest 2026
A git-backed, Markdown-native blackbox flight recorder and skill auditor for autonomous agents. It governs the layer permissions and orchestration don't — memory and competence. Every action is an immutable decision record; audit heuristics turn uncited decisions into draft skill proposals a human approves. Agents propose; humans promote. Grounds on Microsoft Foundry IQ, kept orthogonal to governed memory.
PREVERA Guardian+AI
V-JEPA 2.1 + 3D LIDAR · sole inventor
USPTO PGA-2026-001 · Patent pending
A dual-path agentic fall-prediction pipeline deployed to NVIDIA Jetson Orin. Fast path: under 200ms fall detection. Prediction path: forecasts pre-fall motion 2–3 seconds before impact via a V-JEPA 2.1 encoder/predictor over BEV depth maps. Sensor fusion with LIDAR, pressure mats, and clinical context. HL7 FHIR R4 integration, fairness-audited across mobility and body-type subgroups.
Provenance
Citation-grounded multimodal RAG · no GPU
Live
Ask a textbook a question; get an answer where every claim is checked against the exact page image that proves it. Retrieval is Cohere Embed v4 over 1,347 page images; answering and verification are Claude vision calls in a LangGraph verify→repair loop. Faithfulness 0.985, recall@5 0.864 — reported as measured, not tuned. Fully API-based and serverless.
Clinical AI Agent
SMART on FHIR + transparent RAG
Live demo
Patient-grounded, citation-traceable clinical decision support on open standards. A 5-agent pipeline fuses a patient's live SMART on FHIR record with cited evidence, emits dual citations (patient data + literature), and scores its own faithfulness with a built-in Clinical Work IQ harness. Built to fill the unoccupied square in a scan of nine frontier healthcare-AI companies: open, transparent, patient-grounded, self-proving.
À la Carte
Design-engineering case study · solo
Live
A reading-group companion app — and a working argument that the typed-component pattern from frontier-lab game-playing research generalizes outside of games. Six recipe cards, five lifecycle states, one canonical card catalog that fills session by session. Built end-to-end in a deliberate bake-off with Claude and Cursor. Demonstrates the AutoHarness / Code-World-Models pattern on a non-clinical surface.
FetchMerck AI
Published RAG · clinical decision support
DOI 10.57967/hf/8101
An end-to-end RAG pipeline published on Hugging Face with a DOI: ingestion, chunking, embeddings, semantic retrieval, re-ranking, prompt engineering, deployment. LLM-as-a-Judge scoring rubrics across multiple clinical query classes — the evaluation infrastructure is the asset, not the model itself.
Also shipped — earlier work & explorations
Calibrated, falsifiable exam-readiness with a 60-second reliability-diagram check. Gradio on Hugging Face — free tier, so it sleeps when idle and wakes in ~25s.
A LICENSE-style AI-USE.md declaring how AI touched a project — plus a CI checker that fails when the declaration goes missing or stale.
Offline-first EEG neurofeedback for the Mind Media NeXus-10: EDF/EDF+ verifier, montage configs, web dashboard. PyPI + live app.
Interactive in-browser Python data-structures tutorial — real Python executes client-side via Pyodide, zero backend. Live on GitHub Pages.
US chronic disease prevalence — diabetes, obesity, heart disease, inactivity — carried from data prep through interactive visualization. Live dashboard.
A 280K-line third-party WiFi-CSI pose codebase I audited: built the eval gate that exposed the flagship pose metric as unmeasured, and the ADR recording it. Fork · 66 ADRs · archived.
Thin client and runnable examples for the Semantic Scholar Academic Graph, Recommendations, and Datasets APIs.
Local-first macOS menu bar + web dashboard for a self-hosted Monero P2Pool rig. Reads your own node and miner; the wallet address never leaves the machine. BSD-3.
Public research & reading-group agent on Hugging Face — free tier, so it sleeps when idle and wakes in ~25s.
Multi-model deliberation (Karpathy-style, 3-stage) as an MCP server. MIT · PyPI.
scikit-learn Random Forest · RMSE 277.28 · R² 0.9326 · model + Streamlit + Flask.
V-JEPA 2 + Claude analyzing medical procedure videos into teaching content.
30-day readmission prediction · XGBoost · SHAP · 0.76 AUC-ROC.
Predictive maintenance for wind turbines · TensorFlow.
Safety-first helmet detection — VGG-16 transfer learning with threshold analysis tuned for Zero-Harm deployment.
Smart-contract security learning log: CTF writeups, contest findings, and vuln-class notes as they land.
Plain-language guide to spotting AI-made photos, video, and voice — and talking about them with your teen. No tech background needed.
From bedside
to build.
The methodology is the same across every chapter: design the system that lets a real person do something hard, with humans in the loop and the stakes visible.
1// patient context, pulled live via SMART on FHIR2GET /Observation?patient=demo-134[o-a1c] HbA1c = 8.2 % -> flag5[o-egfr] eGFR = 52 -> flag6[o-ldl] LDL = 110 mg/dL
Proof, not
promises.
Responsible by
construction.
Fairness, interpretability, and honest evaluation aren't a compliance afterthought. They're design constraints I build in from the first commit — especially for systems that touch patients.
Fairness audits
Subgroup performance checks across mobility and body-type cohorts before anything ships — built into PREVERA Guardian+AI.
Honest evaluation
LLM-as-a-Judge rubrics, a reproducible Clinical Work IQ harness, and scope-and-saturation reporting. Numbers as measured, not tuned.
Interpretability
SHAP attributions, append-only decision records, and dual citations you can verify against the source by eye.
Human-in-the-loop
Agents propose; humans promote. Programmatic labeling and weak supervision with a person in the loop, and revocable memory.
A practitioner's
toolkit.
Verified through the patents, published research, and shipped systems — not résumé decoration.
Built in public.
Runnable, reproducible.
Several systems ship as public, runnable artifacts — MCP servers, Hugging Face Spaces, and DOI-published pipelines. Reproducible from the repo and gated by offline tests.
MCP-native
Two MCP servers — llm-council and Compass — callable from Claude Code.
Reproducible evals
Offline CI gates, golden sets, and honest scorecards in every repo.
Live demos
Runnable Hugging Face Spaces and Vercel deployments, not screenshots.
DOI-published
Citable research — DOI 10.57967/hf/8101.
# multi-model deliberation, from Claude Codeclaude mcp add llm-council \--env OPENROUTER_API_KEY=sk-or-... \-- llm-council-mcp> council_deliberate("ship v1 Friday?")# 3 YES / 1 NO -> chairman synthesis
A non-linear path with through-line.
Every role taught something that compounds in the AI work: clinical reasoning under pressure, mixed-methods research, multi-stakeholder delivery, and consultative communication with expert and non-expert audiences alike.
Technical Founder · PREVERA Guardian+AI
Independent · Remote
Shipped ten AI applications zero-to-one, with one USPTO provisional filed as sole inventor — all on Claude's tool-use API. Full ownership: research, architecture, deployment, evaluation, patent prosecution. Advisory engagement with The Gerontechnology Group on ethical AI for aging populations. Most recently: Governance Drift Researcher (published to PyPI with trusted publishing and a clean-room wheel test) and Triton Kernel Lab. Earlier: À la Carte and Compass-BlackBox IQ (Agents League @ AI Skills Fest 2026).
Special Education Teacher
Hawai'i Department of Education
Designed data-driven experimental interventions for 20+ students per year — mixed-methods research with iterative calibration under tight timelines and real stakes. Coordinated cross-functional IEP teams: the literal lifecycle-state-refinement loop À la Carte's architecture is modeled on.
Associate Director of Admission & Adjunct Faculty
Hawai'i Pacific University
Consultative delivery at scale: understood objectives, designed academic solutions across programs, navigated multi-stakeholder decisions, and drove outcomes against quarterly targets. Taught research methods as adjunct psychology faculty.
Special Education Teacher
Hawai'i Department of Education
First five-year stretch of IEP portfolio delivery across diverse classroom contexts, coordinating cross-functional teams under legally binding targets.
Anesthesia & Psychiatric Technician
Queen's Medical Center · Level I Trauma Center
Eight years of frontline clinical reasoning under uncertainty — perioperative workflows, malignant hyperthermia response, psychiatric stabilization. The domain foundation under every AI system I now build.
The training behind the work.
Graduate research methodology, frontier AI training, and cloud fundamentals — the foundation for every system I ship.
Post Graduate Program in AI/ML
Great Learning · McCombs School of Business · GPA 3.95
Anthropic Academy — 13/13 Courses
Claude API, prompt engineering, agentic systems
AI Developer Path — Agents 101 & 201
Agent architectures on accelerated inference
Azure AZ-900 & AI-900
Cloud fundamentals · ML, computer vision, NLP, generative AI
Advanced MLOps & SQL Analytics
Machine learning operations and SQL analytics badges
Patent Filing — Sole Inventor
PGA-2026-001 · LIDAR + V-JEPA 2.1 fall-prediction architecture
PhD ABD — Integrative Mind-Body Medicine
Classic Grounded Theory methodology
MS Counseling Psychology · BA Psychology
BA Magna Cum Laude
Biofeedback Certification
Biofeedback Certification International Alliance · applied psychophysiology
Let's build something
that matters.
Open to roles at frontier AI labs and healthcare AI companies — remote, or relocating to SF or Seattle. Also open to advisory and consulting at the intersection of clinical AI, evaluation, and human-in-the-loop product design.
Based in Seattle, WA · Honolulu & SF · UTC−7