Skip to content

Stefano Petrina

@stefano_petrina

CS student at LMU Munich

CS student at LMU Munich

Student Research Assistant, Ludwig Maximilian University of Munich

Mauern, Deutschland

I'm Stefano,

5th-semester CS student at LMU Munich (specialising in Computational Linguistics), originally from Italy.

What I bring

  • I learn best by building: when something interests me, I try to implement it myself. That's how I ended up writing a small interpreter in Java and an NLP library in Haskell
  • AI / NLP fundamentals: from classic models (perceptrons, n-grams, word embeddings) to neural architectures I've implemented in coursework (LSTMs, CNNs)
  • Systems understanding: how operating systems manage memory and processes, why Rust's ownership model makes it memory-safe and the basic protections against memory exploits
  • Teamwork: built a full Rust app in a university team project and I work part-time on an LMU learning-analytics tool alongside my studies

What drives me

I'm most motivated when I get to understand what happens beneath the abstractions. Right now that pull is strongest toward security: I'm taking a seminar on software hardening, fuzzing and binary analysis, and I'm planning a master's in cybersecurity. I'm still at the start of that path and I want to go much deeper.

Agentic payments caught my interest for a similar reason. If an AI agent can pay for things on its own, a bug or a manipulated prompt doesn't just give a wrong answer anymore, it costs real money. I find the question of how to make something like that safe really interesting and I'd like to spend a weekend working on it with my team.

Code worth a look

  • BuffBuddies: client-server fitness app written entirely in Rust (Iced), with workout tracking, progress charts and a social/gamification layer (team project)
  • haskell-nlp: NLP library covering n-gram models, query-likelihood ranking, Rocchio feedback and a perceptron classifier, all built on my own linear algebra code. Much of it is really about statistics: probabilistic language models, smoothing and likelihood estimation
  • SimpleInterpreter: tree-walking interpreter for my own small programming language in Java, with tokenizer, recursive-descent parser, AST and a call stack supporting functions and recursion