Projected readinessForecast trend
Today+30 days+60 days
AI / Machine Learning Engineer · Honolulu, Hawaii
Building practical AI systems for vision, data, and automation.
I work across computer vision, backend engineering, forecasting, and model-powered tools—with an emphasis on systems that are useful outside the demo.
View selected workSelected systems
2024 — NOW01COMPUTER VISIONReal-time object tracking
60 FPSlive processing
Live-stream detection with YOLO and YOLO-E, SAM 2.1 segmentation, persistent track IDs, and robust occlusion handling.
PythonOpenCVYOLO-ESAM 2.1
Pipeline
YOLO and YOLO-E detection feed SAM 2.1 segmentation, refining masks and boxes before persistent track IDs are carried across the stream.
Engineering
Built frame capture, object association, overlay rendering, multi-object tracking, and occlusion-handling components in Python and OpenCV.
Research
Evaluated SAMURAI-based workflows to reduce bounding-box drift and preserve continuity in crowded, motion-heavy, partially obstructed scenes.
02APPLIED AIReadiness forecasting
70%+less manual review
A FastAPI system turning PostgreSQL supply data into forecasts, anomaly signals, explainable risk scores, and natural-language analysis.
FastAPIPandasBedrockLangChain
Data flow
PostgreSQL supply data moves through Pandas feature pipelines into forecasting, anomaly detection, and risk-scoring workflows.
Product
Delivered the signals through a FastAPI dashboard that highlights parts at risk of shortage and helps teams prioritize review.
AI interface
Connected Bedrock, LangChain, and MCP tools for natural-language analysis of shortage trends, inventory risk, and forecast explanations.
03MODERNIZATIONLegacy intelligence
100K+lines analyzed
Analyzed large COBOL systems to extract business logic, map dependencies, document risks, and support AI-assisted modernization.
COBOLAWS TransformMCPPython
Scope
Worked across more than 100,000 lines of legacy COBOL as part of an AI-assisted modernization effort.
Analysis
Extracted business logic and mapped dependencies to make a large, interconnected system easier to reason about.
Handoff
Documented migration risks and system relationships to support safer modernization decisions with AWS Transform.
Computer vision, forecasting, agent tooling, and backend systems—built for real operational environments.
My work spans real-time object tracking, supply-readiness forecasting, natural-language analysis, and legacy modernization for Air Force and defense programs.
Illustrative system view · Demo data
From supply signals to readiness decisions.
A representative view of how forecasting, anomaly detection, and risk scoring can surface parts that may affect fleet readiness before they become urgent.
Fleet readiness forecast
60-day outlook · DemoSupply riskComponent mix
HighWatchStable
Risk by subsystemPriority queue
Hydraulics
84
Avionics
68
Propulsion
52
Airframe
31
Shortage horizonEarly warning
Actuator assembly14 daysHigh
Sensor module27 daysWatch
Power unit41 daysWatch
Conceptual interface based on the forecasting workflow described above. Values are illustrative and do not represent operational data.
Model intelligence
THE VISION STACKA closer look at the vision stack.
Real-time computer vision is not one model. It is a chain of decisions—detect the object, understand its shape, preserve its identity, and deliver the result fast enough to matter.
Detection
YOLO · YOLO-E · Ultralytics
Fast object detection for live video: locating targets, producing bounding boxes, and establishing the first signal in a real-time vision pipeline.
Segmentation
SAM 2.1
Pixel-level mask refinement around detected objects, used to improve spatial precision beyond a bounding box and support more useful visual outputs.
Tracking
SAMURAI · custom association
Maintaining identity over time through motion, crowding, and partial obstruction—with a focus on reducing drift and preserving track continuity.
ML foundations
PyTorch · TensorFlow · scikit-learn
The core training, experimentation, and classical machine-learning toolkit behind model evaluation and applied predictive workflows.
Beyond the model
PRODUCTION SYSTEMSThe model is one part of the product.
AI orchestration
Bedrock, LangChain, and MCP for connecting models to tools, structured data, and natural-language analytical workflows.
Data intelligence
Pandas, NumPy, PostgreSQL, forecasting, anomaly detection, feature engineering, and explainable risk scoring.
Production APIs
FastAPI, Flask, Node.js, and REST interfaces that turn model logic into reliable, consumable software.
Delivery
Docker, Kubernetes, S3, SageMaker, Git, and AWS services for packaging, deployment, storage, and iteration.
Automation
Playwright-powered research and data collection workflows, paired with summarization and structured metadata extraction.
Modernization
AWS Transform-assisted legacy analysis: extracting COBOL business logic, mapping dependencies, and surfacing migration risk.
How I approach the workUnderstand · Build · Deliver
Start with the signal
Understand the data, operational constraints, edge cases, and the decision the system actually needs to support.
Compose the pipeline
Choose models and features for the real requirement, then evaluate continuity, failure modes, and explainability—not just a headline metric.
Build the product layer
Wrap intelligence in APIs, interfaces, automation, and monitoring so people can use it reliably in a live workflow.