About

Hi, I’m Cristóbal! MSc in Data Science at Universidad de Chile.

My interests lie in diffusion models and how to adapt them. Specifically, I’m interested in using reinforcement learning to build agents on top of pretrained models and gain control over them to (i) expand their generative capabilities, (ii) add specialization through reward functions (RLHF), and (iii) implement exploration mechanisms for control and potential mitigation of mode collapse. I believe that mastering these areas could lead to powerful abstractions in human-computer interaction, enhancing applications and products such as cinematics, games, and scientific tools.

For my master’s thesis, I focused on understanding the fundamentals of diffusion models, policy gradient methods, and their intersection, under the supervision of Felipe Tobar (GAMES group: Group of MAchine learning, infErence and Signals). I am currently continuing my research in these areas.

During my master’s program, I completed an internship at the Paranal Observatory (ESO), where I spent nearly 100 days collaborating with the software team to research and prototype language model applications for the observatory operations. It was an incredible experience, working in the middle of the driest desert in the world, and close to the stars.

Before my master’s, I spent two years working on a public policy initiative aimed at expanding financial inclusion in Chile’s payment systems. The project, initially a joint venture between public and private funds, focused on offering m-POS payment solutions to small and medium-sized enterprises. I was part of the team that scaled the operation to 90k merchants and USD $11M in revenue. My role spanned multiple areas, from managing operations (transactions, data, payments) to overseeing fraud risk (detecting fraudulent transactions and misuse of the service), as well as analytics. The initiative eventually became the payment solutions subsidiary of Banco Estado.

I began my career in the macroeconomic statistics department at the Central Bank of Chile, where I spent three years focused on developing and tracking macroeconomic indicators for the local debt market using microdata, as well as other data projects.

A bit more about me: I contributed a chapter on digital literacy and computational thinking to the book Inteligencia Artificial y Bienestar de las Juventudes en América Latina (127-135). In my free time, I enjoy reading sci-fi, non-fiction and novels. I don’t know why, but I’ve never been able to rate a book I’ve read with less than 4 stars (of 5).

If you’re interested in collaborations or want to discuss any of the topics above, feel free to send me an email (cristobal at alkzar.cl). My CV is available here.

Timeline

• 2016 - 2019: Central Bank of Chile
• 2019 - 2021: Compraquí (BancoEstado)
• 2022: Start MSc in Data Science
• 2023: European Southern Observatory
• 2023: Started Vendie
• 2024: MSc in Data Science completed

Presentations

"Building LLM-based Systems & Products at Paranal"
Slides - 01/2024 Paranal Observatory - European Southern Observatory

"Sequence Models with Transformers. A Journey Through Language Models, Current Insights, and Practical Applications at Paranal Observatory" .- Camilo Carvajal, Cristóbal Alcázar
Slides - 10/2023 Paranal Observatory - European Southern Observatory

"Visualización de Datos con Python"
GitHub / Open in Colab / Slides / Recording - 10/2021 Women in Data - Guatemala Chapter (Webinar)