I'm a self-taught developer from Argentina. I build offline management software for small businesses in Python and SQLite, and I run small experiments to understand how AI systems behave and where they fail.

I don't have a traditional academic background. I learn what each project needs as I go, and this GitHub is where I share my experiments, including what I got wrong and what I learned from it.

I run Software Boutique, where I build custom desktop applications for small businesses: they work without internet, have no monthly fees, and keep the data on the client's own computer. That code is private because it's client work, so it isn't shown here.

An experiment on how in-context learning guides change the behavior of an open-source model (gpt-oss-20b): 840+ automated tests across six configurations, with response times, accuracy and ANOVA analysis. The raw data is public and the time statistics are reproducible. Published as a preprint on arXiv.

An exploratory red-teaming test developed for OpenAI's gpt-oss-20b Red-Teaming Challenge on Kaggle (2025). In some tests, a symbolic language built from blank spaces sent the model into long generation loops.

An experimental text cipher based on a symbolic language I invented as a kid, built in a few days as a learning exercise. It is not secure and shouldn't be used to protect real data; the repository documents its known weaknesses.

• I design the architecture and use AI as a tool. I define how a system is structured and use AI to help build and test it.
• I learn by building. I explore new fields, like statistics or cryptography, when a project needs them.
• I review my own work. My earlier write-ups were more embellished than the data supported. I'm revising them toward plain, verifiable language, and I value finding where my own work breaks.

This space documents my path, one experiment at a time.