Jorge Roldán

I am a machine learning engineer who loves building safe, robust, and awesome AI-powered tools, applications, and services. I have an M.S. in Computer Science with a focus on machine learning from New York University—Courant Institute of Mathematical Sciences. I also studied Computer Science and Mechanical Engineering at The City College of New York—Grove School of Engineering. I am originally from Santa Rosa de Osos, Colombia 🇨🇴, and currently live in New York City 🗽.

My interests lie at the intersection of applied mathematics, machine learning, and software engineering. I believe that truly understanding mathematical and scientific foundations is essential for building great technology. As an engineer, I strive to balance diving deep into theory with getting my hands dirty by building things I love.

Through this blog I want to share what I care about, what I learn, what I build, what I’d like to see more of in the world, and what I’d like to see less of.

If you’d like to share your thoughts or just say hello, you can reach me at roldanjrg@protonmail.com

Huggingface deep dive: Sequence Classification with BERT

Introduction LLMs (Large Language Models) have revolutionized NLP (Natural Language Processing) and are still transforming the field and its applications as of 2025. These models excel at common NLP tasks such as summarization, question answering, and text generation. A common trend in state-of-the-art LLMs is that they base their architecture on the Transformer’s architecture 1 , and decoder-only models have gained favorability compared to encoder-only or encoder-decoder models 2 . In this article, I will discuss how to use the BERT (Bidirectional Encoder Representations from Transformers) model 3 for a sequence classification task with the Huggingface’s transformers library. Remember that BERT is technically just a language model due to its relatively small sizes (~ 100 to ~350 million, depending on the version) compared to large language models with billions of parameters, and it is an encoder-only model. Nevertheless, as I will argue next, understanding and knowing how to use this model is essential. ...

N-Gram Language Models

This post is based chapter 3 from Speech and Language Processing by Dan Jurafsky and James H. Martin N-Grams models N-gram models are the simplest type of language models. The N-gram term has two meanings. One meaning refers to a sequence of n words, so a 2-gram, and 3-gram are sequences of 2, and 3 words, respectively. The second meaning refers to a probabilistic model that estimates the probability of a word given the n-1 previous words. ...

Setting up a Conda environment

TLDR 1 conda create --name my_conda_env python=3.11 1 conda env list 1 conda activate my_conda_env 1 pip install <package_name> or 1 pip install -r requirements.txt What’s a conda environment? Knowing how to set up a Conda environment is an essential skill for any data scientist or Python developer. Conda is an open source package management system for Python. In this post, I will show you how to set up a Conda environment for your project, doing this, will help you to easily install and use any dependency you will need. ...