ALVIN ALIAS / TWO READINGS
Prologue · two readings
Seattle, WA
Physical readinggeometry / forces / heat
Data readingsensors / degradation / remaining life
Drag the line to read it both ways
Run to failure
RUL 112

A jet engine, read two ways: the CMAPSS Turbofan from Reading 02, where RUL means remaining useful life.

I read systems once as physics, once as data.

I spent twelve years in oil and gas, subsea, and HVAC manufacturing. Now I'm finishing an MS in Data Science at the University of Washington and building models that read those systems as data. Ten projects have live links. Six let you run the model in the browser.

The range
Skill galaxy

A career that changed phase.

Twelve years in engineering gave me a real feel for how systems work. Reliability, simulation, cost, manufacturing limits, and tradeoffs. Data science is the expansion. It turns that domain knowledge into features, models, and APIs, and the same methods now run on retail returns, grid demand, and federal contracts.

Skills drawn as a change of phase: an engineering foundation, a domain-feature-engineering capillary, then a data-science fieldEngineering skills gather in a dense foundation on the left, funnel through a long narrow capillary where domain feature engineering turns engineering knowledge into model features, then expand into a wide field of data-science skills on the right.engineering foundationsystems · reliability · deliverydomain feature engineeringwhere engineering becomes featuresdata science rangemodels · inference · deploymentPlant documentationFEA & simulationHVAC systemsReliabilityBOM & supplier costGradient boostingSHAPTime-series forecastingCausal inferenceNLPFastAPIAnomaly detectionA/B testingPyTorch · buildingLLM & agents · buildingCloud ML (GCP) · buildingMLflowDockerPythonSQLscikit-learn12 years
The shape is a change of phase, the same physics that makes air conditioning possible, drawn as my career. Read the story
The full stack
Machine learning10 skills Statistics & inference4 skills MLOps & deployment6 skills Programming &data tools3 skills Engineering domainmanufacturing & reliability7 skills Leadership& delivery4 skills Domain feature engineering Gradient boosting SHAP Anomaly detection Cost-sensitive classification Segmentation & clustering NLP RAG & retrieval evaluation PyTorch LLM & agents
branch selected
Machine learning

10 skills are open on this branch. Evidence appears where the branch is brushed.

Machine learning 10 skills
Statistics & inference 4 skills
MLOps & deployment 6 skills
Programming & data tools 3 skills
Engineering domain: manufacturing & reliability 7 skills
Leadership & delivery 4 skills
Ten shipped systems

Each project is one system, read twice.

Down the left is the index: every system I have shipped, each tagged with a defensible metric or scope marker. Hover or tap any line to open its reading on the right, then click to keep it open while you follow the links. The top half is the physical reading, what the system is and the engineering call I made. The bottom half is the data reading, the metric and one honest limit. Same system, both halves.

The index The reading
Reading 01retail / ops
1M+ real transactions

Retail Returns Intelligence

A retail analytics prototype spanning customer anomaly flags, segmentation, substitute recommendations, and a policy simulation. A later audit found leakage in the predictive classifier and backtest, so those performance claims are quarantined while the target and features are rebuilt.

1M+rows audited
The strongest result here is the audit: a temporal split does not rescue features that encode future history or cancellation signs.
LightGBMIsolation ForestDuckDBFastAPI
Honest limitI do not present the retained 0.992 ROC-AUC or 8.7x lift as forward-prediction evidence. The rebuild needs purchase-time labels, point-in-time features, and a new untouched evaluation.
Case studyLive demoSourceAPI
The throughline

Twelve years of engineering is the domain depth my models run on.

Most data scientists start with the data. I started twelve years earlier, with the machines that produce it, and worked my way to the models. That order is the point: real BOM and field-failure data at Daikin ($5M+ found), dashboards and an ML push at Rheem, and the formal toolkit on top.

EducationFirst-Class Honors · Government Engineering College, Thrissur

The mechanical foundation: physics and design.

ExperienceMechanical Engineer, via KH Consulting · Baker Hughes, Houston TX

Turned field measurements into plant documentation and a structured dataset to replicate a plant overseas. The first sight of industrial data at scale.

ExperienceDesign Engineer · Centurion Subsea Services, Katy TX

High-pressure subsea tooling where failure is unrecoverable. The failure-mode intuition behind how I model risk today.

ExperienceDesign Engineer III · Daikin North America, Waller TX

Analyzed BOM and supplier datasets across HVAC product lines to surface $5M+ in annual cost reductions, and mined field and production data for the root causes behind reliability failures. The closest thing to data science before it had the name.

ExperienceSenior Project Engineer · Rheem Manufacturing, Fort Smith AR

Built dashboards that sped up how engineers read lab-test data, and made the internal case for training ML on years of past test data to sharpen in-house simulation. Led the 2023 DoE and 2025 refrigerant redesigns, and cut a simulation test cycle by 30%.

EducationM.S. Data Science · University of Washington, Seattle

Statistics, machine learning, experimental design, data management, and software design. Expected March 2027.

End of readings

The best calls come from people who can read the physical system and the data describing it. I spent twelve years on the first half, and I am all in on the second.

alvin.alias@gmail.com LinkedIn GitHub Resume on request

Targeting full-time data science roles where domain depth and shipped work matter. Graduating March 2027, open to conversations now.

Like the night sky? Fly through my skills →
alvinalias.com Two readings. One engineer.