Rubén
Science begins with curiosity

Hi, I'm Rubén Cañadas

My name is Rubén, and to be honest, I don't really know what I am.

I'm not a physicist.

I'm not a mathematician.

I'm not an AI engineer.

I'm not a software engineer.

Or maybe I'm a bit of all of them.

I like everything, I learn from everything, and I refuse to put myself in a box.

On a good day, we can talk about pseudo-Riemannian manifolds or tensor algebra.
On another, about how quantizing the Poisson bracket gives rise to commutators in quantum mechanics.
And if that feels too theoretical, we can jump straight into improving a multi-head attention mechanism or fine-tuning a Transformer with QLoRA.

Curiosity is the common thread.

That's why I've decided not to define myself on my own website.
I'm not a title. I'm not a label.

I'm just someone who's endlessly curious.

I could tell you about my career path, my degrees, my experience — all the usual stuff.
It would be perfectly fine.

But everyone does that.

That's what LinkedIn is for.

But anyway, here's what I've been focused on lately:

I'm bringing physics, mathematics, and artificial intelligence together in the world of computational biology.

The goal is not just to simulate molecules or predict structures.

It is to understand, in depth, how biology encodes information.

How physical laws, mathematical structure, and chemical constraints shape molecules and proteins.

How meaning emerges from form.

How function is written into matter.

Hidden inside molecular structures is an intrinsic language — a code shaped by evolution and physics.

By learning to read that code, we can build models that don't just predict outcomes, but understand.

Systems that capture structure, dynamics, and intent.

Systems that reason about biology, rather than approximate it.

That is the direction I'm exploring.

Currently Learning

Things I'm Exploring Right Now

Topics and technologies I'm currently diving deep into

CUDA Kernels with Triton

Writing custom CUDA kernels using Triton to optimize inference, finetuning, and training pipelines. Achieving significant speedups in deep learning workloads.

CUDATritonPyTorchOptimization

Algebraic Topology for Molecular Docking

Exploring how topological data analysis and algebraic topology can help find optimal binding spaces for molecular docking simulations.

TDAPersistent HomologyDrug Discovery

Preference Alignment (DPO, RLHF, RLAIF)

Studying and implementing preference alignment techniques like Direct Preference Optimization (DPO), RLHF, and RLAIF to better align language models with human preferences.

DPORLHFRLAIFLLM Alignment
Open Source

Open source projects

Community contributions that aim to solve real problems and accelerate research.

MolFun

High-performance library to accelerate training and inference calculations for molecular modelling models like Protenix, AlphaFold and more.

PythonCUDAPyTorchNumPy

10x faster

Low-level optimizations for molecular calculations

GPU Accelerated

Native support for CUDA and parallel operations

Easy integration

Compatible with PyTorch, JAX and popular frameworks

Molecular Modelling

Optimized for Protenix, AlphaFold, ESMFold

Stay Updated

Get notified about new posts and projects

No spam, unsubscribe anytime. Your email is safe with me.

Contact

Let's Connect

Have a project in mind, a question about my work, or just want to say hello? I'd love to hear from you.

Get in Touch

Have a question or want to work together?

0/5000 characters