About

I'm Max. I'm passionate about AI — all of it. From classical ML and reinforcement learning to today's generative models, the whole arc is what excites me.

What I'm really chasing: making AI accessible in Latin America, and helping it feel like something of ours — built here, in our languages, on our problems. AI as part of our identity, not borrowed.

Currently AI Engineer at Cashea . Python, C++, CUDA, TypeScript.

Now

  • → Learning CUDA from first principles — writing kernels by hand to see what the compiler does for you, and what it doesn't.
  • → Implementing FlashAttention from scratch to understand IO-aware algorithms, then profiling against the real thing.
  • → Reading PTX output and Nsight traces — figuring out where the cycles actually go during LLM inference.
  • → Building EigenCore — making AI accessible and sovereign in Latin America.

Writing

De la física a los LLMs

es

Cómo pasé de resolver ecuaciones en un pizarrón a entrenar modelos de lenguaje — y qué me llevé en el camino.

all posts →

Work

  • AI Engineer at Cashea

    Building AI agents.

  • ML Engineer at BBVA

    Building ML models.

Projects

  • eigencore eigencore.ai ↗

    A LATAM-first AI company. Building accessible, sovereign artificial intelligence — trained here, on our languages, for our problems.

  • Tlama 124M github.com/eigencore/Tlama_124M ↗

    An open-source Mexican LLM. A 124M-parameter language model, pre-trained from scratch. A proof that Latin American teams can build foundation models from the ground up — not just fine-tune the ones we import.

Talks

  • ¿Qué es Tlama? El LLM mexicano open-source
    AMPLN — Asociación Mexicana de PLN

    Seminar talk on building Tlama 124M from scratch — motivation, architecture, training infrastructure, and what it means to ship a foundation model from Mexico.

Contact

The fastest way to reach me is email . I'm also on X and GitHub .