Currently engaged Oregon Department of Education Business Intelligence Analyst · Since Dec 8, 2025

Epsilon Labs · Independent data science practice

Data into decisions you can defend.

I’m Brian Cervantes Alvarez — a statistician who builds. Epsilon Labs is my one-person practice: statistics-first, AI-native, and in service of the community organizations and teams around me who need answers that hold up under scrutiny.

See the work Start with the basics

Contract work: currently unavailable — paused while I serve at ODE. Community questions are always welcome.

The discipline

What is data science?

It's the craft of turning recorded facts — enrollment counts, sensor readings, survey answers — into answers you can act on. Done well, it is less about algorithms and more about asking precise questions, measuring honestly, and knowing exactly how much to trust the result.

Ask & collect

Define the question precisely, then gather data that can actually answer it. Most projects are won or lost right here, before any model exists.
Model & test

Fit statistical and machine-learning models — then try hard to prove them wrong. A model you haven’t attacked is a model you can’t trust.
Decide & monitor

Translate results into a decision, ship the tooling, and watch whether reality agrees. The analysis isn’t done when the report is.

The stakes

Because guessing is expensive.

Every organization already makes data decisions — the only question is whether the data gets examined carefully or just gestured at. Three habits separate the two:

Averages hide the story

Two programs with the same average outcome can serve completely different people. Distributions — not summaries — are what good decisions are made of.

Uncertainty is information

A forecast without an error bar is an opinion. Quantified uncertainty tells you when to act, when to wait, and when to collect more data.

Reproducibility compounds

An analysis that reruns cleanly next quarter becomes infrastructure. One that doesn’t becomes folklore. I build the first kind.

The practice

From open question to defensible decision.

All services

Four practice areas, one throughline: every engagement ends with something your team can run without me in the room.

Statistical analysis & modeling

Experimental design, hypothesis testing, regression, and Bayesian inference for questions that deserve careful answers.

Predictive modeling & ML

Classification, forecasting, and risk scoring with proper validation. Models built to hold up under audit.

Interactive dashboards

Self-service decision tools your team can poke at without me in the room — defaults that surface signal first.

Reproducible reporting

Pipelines that document their assumptions, version cleanly, and produce the same numbers next quarter.

AI-native

Models as collaborators, not oracles.

Frontier models are part of how I work every day — for code, for drafts, for analysis pairing. The craft is knowing when they're confident and when they're guessing: I write the prompts, run the evals, and validate output against ground truth before anything ships.

ChatGPT

Reasoning, drafting, code review at speed.

Claude

Long-context analysis, careful writing, hard problems.

Gemini

Multimodal exploration, large doc ingestion, fast iteration.

Automation with receipts: anything a model touches gets the same validation discipline as anything I write by hand.

Proof

Selected projects.

All projects

Dashboards, models, and analyses built end-to-end — each one links to the full write-up with methods and code.

Dashboard

UK Traffic Accident Trends

An interactive dashboard for exploring incident patterns — filters, dynamic views, defaults that surface signal first.

Statistical modeling

Multivariate Analysis: Wine Quality

PCA and supervised models with diagnostics to separate signal from noise across correlated chemical features.

Geospatial dashboard

Yelp Business Reviews

A spatial review explorer with map-linked filters, density layers, and per-business summaries.

On stage

Talks & workshops.

Full archive

Live teaching is where the practice sharpens. Every deck below is embedded on its page — open one and present it fullscreen.

Brian Cervantes Alvarez

M.S. Statistics, Oregon State · M.S. Data Science, Willamette

R & webR
Quarto
Applied statistics
Teaching

Open shelves

The resources I actually use.

Full list

Free, canonical, and community-maintained — the same shelves I point students and clients to. No affiliate links, no gates.

Tools & publishing

Community & reading

Foundations

Open data

My own shelf lives on the blog — teaching labs, visualization walkthroughs, and applied write-ups.

An honest invitation

Rooted in community, not a startup.

Epsilon Labs looks like a modern studio, but it's one statistician serving his community. Right now my full attention belongs to the Oregon Department of Education, so contract engagements are paused — but the door isn't closed.

Have a question worth answering carefully?

Community organizations, educators, students: send the question and the data you have. I'll reply within two business days with next steps — or point you to someone who can take it on if I can't.

Backed by graduate training in Statistics (M.S., Oregon State) and Data Science (M.S., Willamette) — the foundations live underneath every model and every report.

Email me About Resume