Erick Galván

Studies

Education

High Scool

I studied at CBTIS 168 "Francisco I Madero", in Aguascalientes, Mexico from 2016 to 2019 where I graduated with a technical degree in Programming

At first I was not sure about studying programming, to be honest, when I started I didn't like it at all, maybe because I didn't see the real use of it at the moment. Despite that, when I graduated I already had the bases so I decided to keep studying it.

University

In 2019 I started studying "Ingeniería en Computación Inteligente" (traslated to English it would be something like Intelligent Computing Engineering) at Universidad Autónoma de Aguascalientes. That career is abut programming, focused on artificial intelligence and machine learning.

These were 5 years of a lot of learning. When I was in the 3rd semester I started working at a little company called CCEO Software Development, where I worked as a mobile developer. I was developing mobile applications using Flutter and Dart. That was my first real experience as a developer, so there I learned a lot of things that I didn't learn at the university.

I graduated in 2024.

Thesis

"Comparación cualitativa y cuantitativa de modelos de lenguaje pre entrenados" (Qualitative and quantitative comparison of pre-trained language models) was the name of my thesis. In this thesis I compared different pre-trained language models like GPT, Flan y Mistral. I wanted to know which one was better is general and what was the reason of that.

I used different metrics to compare them, like BLEU, ROUGE, BERTScore, and WikiSplit. I also used some qualitative metrics like the quality of the generated text and the coherence of the text.

To compare them, I created a web application where the user could upload a PDF file and keep a conversation with the model. Based on the conversation, the application would show the metrics of the model, while the user could test the quality of the generated text.

You can see the code of the application here


In general, I found that actually there is not a model that is better, just the one that fits better to the task that you want to do. And based on that, I explained the main factors that you should consider when you are choosing the characteristics of the model that you want to use or build.


if you want to read the thesis, you can do it here