Isometric city drawn in thin ink linework, with a single orange evacuation route threading through the streets to a circled point

MYRON LAI

Maker Portfolio · 2026

  1. 01 HRV-ML Monitor ML cardiac monitoring for SCAD patients, built on the decoupling index. 05–06
  2. 02 HRV-ML Monitor Leave-one-subject-out accuracy, and what the model actually weighs. 07–08
  3. 03 EmergencyPath Disaster response: fast community reports fused with trusted official feeds. 09–10
  4. 04 EmergencyPath Two years from DisasterScope v1 to EmergencyPath v2, measured in users served. 11–12
  5. 05 AI Helmet Patented 360° hazard detection and AR navigation for bikers. 13–14
  6. 06 AI Helmet The intellectual property behind the 360° sensing and AR overlay. 15–16
  7. 07 Systems Two solo builds: an autonomous agent and a compound-risk model. 17–18
  8. 08 Leadership & Dead Ends Teams led, and two dead ends kept. 19–20
  9. 09 Colophon Everything else, and how this book was made. 21–22

Every project links to a live deployment or public repository.

Proofgithub.com/bdbb ↗ github.com/bdbbmyrondomain.com ↗ myrondomain.com
MYRON LAI CONTENTS 03–04

01 / HRV-ML MONITOR

Cardiac early warning from heart-rate variability.

FIG. 01The decoupling index
SIGNAL EXERTION → HEART RATE HRV (RMSSD) DECOUPLING INDEX carries 90.3% of the model
Polar H10 chest rig + Android app in use FIG. 02

Wearable ML system for SCAD patients using heart-rate variability. The discovery: when heart rate rises but variability fails to follow, the gap itself is the signal. That gap became the decoupling index.

Year
2025–26
Role
Solo: research, hardware rig, Android app, ML
Stack
Python · ML · Android · ONNX · Polar H10 ECG
Proofdemo video ↗ www.youtube.com/watch?v=CzkY21VTrtk
MYRON LAI HRV-ML MONITOR 05–06

RIGOR

95.6%
LOSO accuracy, 22 subjects
90.3%
feature importance carried by the decoupling index
39%
mean RMSSD suppression during running (d = 0.90)
99.0%
accuracy with cardiac + accelerometer features
FIG. 03What the model actually weighs
IMPORTANCE → DECOUPLING INDEX 90.3% RESTING HR 4.8% SDNN 3.1% AGE 1.8%
subject mid-run wearing the rig FIG. 04 LACSEF 2026 · APPLIED COMPUTATIONAL SCIENCE 1ST ISEF NOMINEE
HRV-ML demo video still
watch: www.youtube.com/watch?v=CzkY21VTrtk
MYRON LAI HRV-ML MONITOR 07–08

02 / EMERGENCYPATH

Real-time disaster alerts and verified evacuation routes.

FIG. 05Two sources, one confidence score
COMMUNITY REPORTS FAST OFFICIAL FEEDS TRUSTED AI CONFIDENCE SCORING LIVE HAZARD MAP SAFE EVACUATION ROUTES
Live hazard-map dashboard with a safe evacuation route FIG. 06

AI-driven disaster response platform: community reports fused with official feeds under AI confidence scoring, producing live hazard maps and safe evacuation routes. Evolved from DisasterScope (v1) to EmergencyPath (v2).

Year
2024–26
Role
Lead: full-stack, AI pipeline, deployment
Stack
FastAPI · Supabase · Python · AI
Prooflive app (v2) ↗ pa.myrondomain.comv1 ↗ ds.myrondomain.com
MYRON LAI EMERGENCYPATH 09–10

IMPACT

16,900+
alerts tracked
1,500+
visits
250+
registered users
WAICY GLOBAL · OF 130,000 STUDENTS TOP 5
DisasterScope v1 demo: Evacuation Hub title card
DISASTERSCOPE V1 / DEMO
EmergencyPath v2 live map with active alerts
EMERGENCYPATH V2 / LIVE
EmergencyPath demo video still
watch: www.youtube.com/watch?v=Rz-S4dT06y4
MYRON LAI EMERGENCYPATH 11–12

03 / AI HELMET

Patented 360° hazard detection for cyclists.

FIG. 07360° detection cone → rider HUD
HAZARD AR OVERLAY → RIDER HUD
prototype helmet held in hands, sensing ring visualized FIG. 08

Patented embedded AI system for real-time 360° hazard detection and AR navigation overlays for bikers.

Year
2024–26
Role
Inventor: embedded vision, AR overlay, patent filings
Stack
Computer Vision · Edge AI · Embedded · AR
360° real-time hazard detection
3 patents: 1 granted, 2 pending
MYRON LAI AI HELMET 13–14

Intellectual Property

Taiwan Utility Model No. M676830: GRANTED
US 63/860,723: pendingUS 63/858,337: pending

The filings protect the embedded sensing pipeline and the AR overlay that turns detection into a rider-facing heads-up display.

helmet detail close-up: camera array and AR optics FIG. 09

The sensing array up close: embedded vision under a rider's shell.

MYRON LAI AI HELMET 15–16

04 / ENGRAM

FIG. 10Dispatcher → worker pool, with persistent memory
TELEGRAM DISPATCHER MAIN SESSION API ~700ms INTERACTIVE ~25s PIPE ~2s RUNS OUTSIDE THE CONTEXT WINDOW VECTOR MEMORY 60K+ EDGES
60K+
knowledge-graph edges across 750+ items
6+
concurrent workers

Autonomous personal AI assistant on Telegram: a dispatcher session routes tasks to a tmux worker pool (3 executor types, ~700ms–25s startup), with hybrid dense+sparse vector memory and an auto-computed knowledge graph.

Year
2025–26
Role
Solo: architecture, infra, memory system
Stack
TypeScript · Python · Bun · SQLite · MCP · Vector Search

05 / RISKMESH

compound-risk map: wildfire, drought, and grid failure converging on one coastal town FIG. 11
4
federal data sources fused live

Compound disaster risk prediction for Southern California: wildfire, drought, and grid-failure interactions modeled with a stacked XGBoost ensemble over NASA FIRMS, NOAA, USGS, and EIA data, with SHAP explainability.

Year
2025–26
Role
Solo: data fusion, modeling, deployment
Stack
Python · XGBoost · SHAP · FastAPI · PostgreSQL
Prooflive app ↗ riskmesh.myrondomain.com
MYRON LAI SYSTEMS 17–18

Leadership

  • FTC Robotics: Team Captain 2023–25

    Two seasons: rookie-year ILT #1; 2024–25 ILT & League #1 with Regional Qualification. Dean’s List Semi-finalist 2025. Solo programmer for autonomous + driver control.

  • FIRST LEGO League: Coach & Mentor 2023–25

    Coached teams to World Competition (Core Values Award) and West Edge International; translator for international teams. PVSA Gold (100+ hours).

  • AI Club: Founder & Co-Leader 2025–

    Founded the school’s AI club: generative AI, safety, alignment, hands-on projects.

  • Cybersecurity Club: Co-Leader 2025–

    Workshops and CTF practice; led the team to a #12 US Middle/High School finish in picoCTF 2026.

Kept: the Dead Ends

DisasterScope's first backend

v1 ran on Flask and Firebase, deployed like any class project. Under real users it was slow. Requests queued, queries got chatty, and the database was a poor fit for live geospatial alerts.

I rewrote the backend on FastAPI and Supabase, then moved hosting to a VPS swarm with backups so the service stays up during the emergencies it exists for.

FLASK + FIREBASE FASTAPI + SUPABASE VPS SWARM + BACKUPS

HRV before the chest strap

Blood-pressure monitoring first, then optical wrist sensors, including a Xiaomi watch. Cuffs cannot sample continuously, and wrist PPG dropped beats during movement, exactly the moments the data mattered most.

A chest-worn ECG finally gave clean RR intervals, and those made the decoupling index findable. Pick the sensor for the signal, not for convenience.

BP CUFF WRIST PPG CHEST ECG
MYRON LAI LEADERSHIP 19–20

Also Built

  • audit.log Startup: regex + AI codebase auditing for leaked keys and compliance. GitHub + Slack integrated. audit.myrondomain.com
  • Study Platform Adaptive AI learning paths. FastAPI, PostgreSQL, Qdrant. AGI Honorable Roll 2025. studyplatform.myrondomain.com
  • Blackbaud SDK Reverse-engineered school-LMS APIs: async Python SDK, MCP server, Todoist AI sync. 41 unit tests.
  • Conrad Microneedle Depression-biomarker wearable patch concept. 2025–26 Conrad Innovator, Space Center Houston.

One codebase builds this book as both a website and a PDF.

This book, by the numbers

22pages
1typeface
1accent color
7figures drawn as SVG
5automated checks
2outputs, one codebase
Proofgithub.com/bdbb ↗ github.com/bdbbmyrondomain.com ↗ myrondomain.comwww.linkedin.com/in/myronl ↗ www.linkedin.com/in/myronl