> Cynthia

Cynthia Alvarado

San Francisco Bay Area

I'm a builder by disposition, happiest once I've solved a problem and then built the thing that keeps it solved. My family and I traded Texas for the Bay Area not long ago; most evenings I'm cooking dinner, most mornings I'm on a trail or in open water, and there's almost always a Raspberry Pi in the next room doing something it wasn't strictly asked to.

Career

How I got here

I notice patterns. It's less a skill than the thing my mind does on its own, and for a long time it was just how I made decisions, a quiet habit running in the background. Then a high school statistics class gave the habit a name and a method: the same teacher who'd walked me through algebra and geometry handed me statistics, and the patterns I'd been sensing became things I could quantify, test, and use to predict what came next. The instinct had turned into a discipline.

Statistics gave me the framework; programming gave it reach. Data science was where the two met, the place where statistical thinking, code, and real business context could be pointed at questions that didn't have clean answers. That's the work I kept choosing: not the tidy problems but the messy ones, where the data was incomplete, the requirements moved, and the right answer had to be built rather than looked up.

Fifteen years later, that's still the thread. I've built fraud models in consumer lending, written data governance rules at telecom scale, and shipped risk models that steered infrastructure decisions worth millions. Now, at HFD, it's the credit risk scorecards and loss projections that keep a lending portfolio honest. The domains kept changing; the way of thinking didn't. Somewhere in there it hardened into a conviction I co-authored a paper about in 2018: that data about people belongs to those people, and that the models we build on it should be able to explain themselves. It's why I care as much about whether a model is governed and trustworthy as whether it's accurate, and why, off the clock, everything I build runs on hardware I own.

Focus

What I focus on

Credit risk modeling

  • Logistic regression & gradient boosting
  • Scorecards · WoE / IV · PDO
  • PD / LGD / EAD models
  • Validation · PSI · KS · ROC / AUC
  • CECL loss projection
  • Model governance (SR 11-7)

Machine learning

  • Supervised & unsupervised learning
  • Predictive modeling & forecasting
  • Feature engineering & validation
  • Model monitoring & drift detection
  • scikit-learn · TensorFlow · PyTorch

Data, platforms & delivery

  • Python · SQL · Apache Spark
  • BigQuery · Google Cloud Platform
  • ETL & production pipelines
  • Dashboards (Looker)
  • Self-hosting: Docker, Pi, Gitea, Tailscale

Education

M.S. in Data Science, Southern Methodist University, 2018

B.S. in Statistics, University of Texas at Dallas, 2009

Selected work

Personal Projects

Personal

Away from the desk

Away from screens

Most of my time off is outdoors with my two daughters. I'm often on the trails either running or hiking with my daugghters and we swim open water whenever the water cooperates. I cook most nights too, though "cook" may overstate it; my daughter maintains it beats takeout, and I've decided to take her at her word.

Precision instruments

I'm partial to a well-made tool. Fountain pens are the obvious one, and the reason there's iron gall blue-black running through this whole page. The less obvious one is my Wüsthof chef's knife, which does most of the real work behind that dinner framework. Both come down to the same small pleasure: something built with care, kept sharp, doing exactly what it's for.

Contact

Get in touch

Find me on LinkedIn, or write to .

Colophon

corDeHierro: an iron heart. Hierro is iron, the ink is iron gall, the φ is the nib it's written with. Written by hand in HTML and CSS, with no framework, no build step, no analytics, no trackers, and nothing loaded from anyone else's server. Set in Iowan Old Style with a monospace counterpoint, iron gall blue-black on laid paper, inverted after dark. One file, about 28 KB.