From audio engineering to AI systems — 18 months of daily building, and here's what it produced.
I build ML systems, RAG pipelines, and cloud-deployed data tools on AWS and Azure. The projects on this site are working, deployed systems — not exercises, not tutorials. Here's how I got here.
Open to AI engineering and MLOps roles — Madrid or remote. Available immediately.
My Story
How I got here
Before
Audio Engineer — SAE Institute Madrid
Audio engineering is a precision discipline. You're working with imperfect material under deadline pressure, and the output either holds together or it doesn't. I was good at it — but the ceiling arrived fast. I kept running into the edges of what the work could ask of me, and I wanted problems with more depth than a mixing desk could offer.
Pivot
Python & Algorithms Training
Python was the thing that changed the direction. It wasn't gradual — something clicked quickly and made it obvious this was where I was going. The structure of the language matched how I think, and the problems you could solve with it had no ceiling I could see.
The transition was deliberate. Formal Python and algorithms training from scratch, while teaching English to pay the bills. Not a side project — a full switch, taken seriously from the start.
Going deeper
AI, Machine Learning, Data Science & SQL
750+ hours of structured training across AI, machine learning, data science, and SQL — stacked on top of coding three to four hours every day. TensorFlow and Keras for deep learning. scikit-learn for classical ML. The full pipeline from raw data to a model that serves predictions through an API.
The training gave me foundations. The projects are where the actual understanding came from.
Now
Building in production
The work on this site is the result. Not submitted exercises — deployed systems on real infrastructure. A SageMaker pipeline. A serverless API on AWS Lambda. Five applications on Azure App Service. A RAG assistant that anyone visiting this site can actually use.
The habits audio engineering built — precision, the ability to work cleanly under pressure, knowing when something isn't good enough yet — have transferred more directly than I expected.
Where I am now
Over the past 18 months the work has covered the ML stack end to end: data cleaning and analysis, training and evaluation, explainability, deployment, and monitoring. The projects aren't isolated — they form a deliberate progression.
Amazon Bedrock benchmarked across four different task types, with a consistent, published conclusion about where foundation models outperform trained classifiers and where they don't. A full SageMaker MLOps pipeline — Training Job, Model Registry, Real-Time Endpoint — with an Azure App Service frontend calling the AWS model endpoint. A production RAG assistant using LangChain, ChromaDB, and Groq, deployed and accessible directly from this site.
None of this came from a course. It required figuring out Docker, ECR, Lambda Web Adapter, multi-cloud authentication, and SageMaker's training container conventions from scratch. That's what the GitHub repos show.
Where I am going
I'm targeting AI engineering and MLOps roles where the job is actually building and deploying systems — working with foundation models, cloud infrastructure, and real data to produce something that runs in production.
The background is non-standard, and I think that's worth something. Eighteen months of self-directed, daily work produces a different kind of confidence than coursework alone — you know what you can figure out, because you've already had to. The projects on this site are the evidence. The GitHub repos are the detail.
Want to get in touch?
Open to AI engineering, MLOps, and data science roles — Madrid or remote.