01 · The instruction before you even look at results
Entrenalo supervisado.
Train it as supervised.
Voseo shorthand for “aprendizaje supervisado” — this is how it gets said in standup, not the textbook phrase.
andbefluent.com/spanish/data-scientist
Data Scientist
Explain models, neural networks and data pipelines to a Spanish-speaking team — out loud.
The core paradigms translate cleanly: el aprendizaje supervisado (supervised learning), el aprendizaje no supervisado, la clasificación and la regresión. Neural networks bring la red neuronal convolucional, la capa oculta (hidden layer) and la retropropagación — though working data scientists just say backprop. The register trick to know: real Spanish-speaking ML teams keep machine learning, data lake and feature engineering in English, say la data for their datasets, but reach for precise Spanish when it matters — el sobreajuste (overfitting), la validación cruzada, el sesgo del modelo (model bias). That code-switching only becomes natural by speaking it, not drilling flashcards.
Below: the vocabulary lesson by lesson — from ML fundamentals to neural nets, training, big-data infrastructure and AI ethics — how teams actually talk across Latin America, and a way to rehearse a technical meeting out loud.
The warm-up pack
The Spanish an ML team actually speaks — half technical, half borrowed straight from English.
01 · The instruction before you even look at results
Entrenalo supervisado.
Train it as supervised.
Voseo shorthand for “aprendizaje supervisado” — this is how it gets said in standup, not the textbook phrase.
02 · What you ask for when the model's stuck
Agregale una capa.
Add it a layer.
Casual for “capa oculta” — you don't need the formal term to ask for another hidden layer.
03 · The verdict nobody wants to deliver
El modelo se sobreajustó feo.
The model overfit badly.
“Sobreajustar” behaves like any Spanish verb here — this is exactly how you'd report bad news in a review.
04 · The warning before you touch the dial
No overfitees el modelo.
Don't overfit the model.
A Spanish sentence built on an English verb, voseo negative imperative — completely normal in these teams.
05 · What you ask for before anything ships
Armá la pipeline.
Build the pipeline.
“La pipeline” stays English inside an otherwise Spanish sentence — that code-switch is the actual C2 skill.
06 · The step before the model sees anything
Limpiá la data.
Clean the data.
“La data” — feminine, borrowed — is what gets said far more often than “los datos.”
07 · The honest read on a biased model
El modelo está sesgado.
The model is biased.
Plain Spanish for flagging it. You'll also hear “tiene bias” in the exact same sentence.
08 · The question a teammate asks casually
Vos hacés NLP, ¿no?
You do NLP, right?
The acronym stays English even inside a fully Argentine, voseo sentence — that's normal, not sloppy.
09 · The three letters that never got translated
CNN se dice en inglés.
CNN gets said in English.
The architecture name stays English even mid-sentence — there isn't a Spanish version anyone actually uses.
10 · The word that replaced ‘la variable objetivo’
Le decimos el target, no la variable objetivo.
We just call it the target, not la variable objetivo.
The formal term for what you're predicting almost never gets said out loud. “El target” won.
11 · What never gets translated in a pipeline
Batch y streaming se quedan en inglés.
Batch and streaming just stay in English.
The processing modes keep their English names even in Spanish-speaking teams — “procesamiento por lotes” lives only on paper.
FLUENT
Finally, someone to talk to in Spanish.
Ten-minute spoken lessons. A teacher who answers back — and gently fixes what you got wrong.
First 10 lessons freeandbefluent.com/spanish/data-scientistRegional Spanish
Textbooks teach one word. Locals use several — pick your region's and stay consistent.
Watch out
The part no drill site can do
Olivia
Your vocabulary teacher for this pack
There are no vocab lists to memorize here — in the Data Scientist lessons you talk, and Olivia keeps handing you the meetings where this vocabulary lives. One lesson you're in an ML team meeting explaining the difference between aprendizaje supervisado and no supervisado, with real examples of clasificación and agrupamiento. The next is an architecture review: you walk through a red neuronal convolucional — capas, función de activación, retropropagación — while she pushes back like a colleague would. Then a data-engineering briefing: your pipeline de datos, from ingestión and transformación to almacenamiento in a lago de datos — all out loud, until the terms come without translating.
Blank mid-sentence and nothing bad happens — she waits. That's the practice, without unnecessary judgement.
Finish the 6 lessons and Data Scientist is yours — earned, not given.
First 10 lessons free · 10-minute spoken lessons · your AI coaching team remembers you
Quick answers
Usually you don't translate it — in technical meetings across Latin America it's machine learning or just ML. The Spanish terms appear when you specify the paradigm: el aprendizaje supervisado, el aprendizaje no supervisado, el aprendizaje por refuerzo (reinforcement learning).
El sobreajuste — and Mexicans use it as a verb: el modelo se sobreajustó. Many teams keep the English: in Argentina you'll hear no overfitees el modelo. Underfitting is el subajuste.
The data lake — with el almacén de datos as the formal term for a data warehouse. In practice both usually stay in English, and everyday jargon calls the data itself la data (feminine, an anglicism) instead of los datos.
La red neuronal convolucional is a convolutional neural network — often just la red or la neural in context, with CNN said in English. The components: la capa oculta, la función de activación, el peso sináptico — though colleagues say los pesos and backprop in daily talk.
Bias is el sesgo del modelo — colloquially el modelo está sesgado, or with the anglicism, tiene bias. Explainability is la explicabilidad del modelo, though in stakeholder talks you'll also hear XAI.