About

My name is Luca. I am passionate about programming languages, security, agentic AI systems and the intersection of these fields. I also really like web technologies like CSS and fancy user interfaces.

During the day, I work on agent security at Invariant. I also created the LMQL project during my PhD at ETH Zürich, a programming language for using large language models (LLMs) in a safe and efficient way.

Publications

During my PhD, I also published a number of papers on LLMs together with a number of great collaborators. Here is a list of some of them:

2024

• AgentDojo: A Dynamic Environment to Evaluate Prompt Injection Attacks and Defenses for LLM Agents (NeurIPS 2024)
  Edoardo Debenedetti, Jie Zhang, Mislav Balunovic, Luca Beurer-Kellner, Marc Fischer, Florian Tramèr
  [Website]
• AI Agents with Formal Security Guarantees (ICML 2024, Next Generation of AI Safety Workshop)
  Mislav Balunović, Luca Beurer-Kellner, Marc Fischer
  
• Prompt Sketching for Large Language Models (ICML 2024) Luca Beurer-Kellner, Mark Niklas Müller, Marc Fischer, Martin Vechev
  [Code]
• Guiding LLMs The Right Way: Fast, Non-Invasive Constrained Generation (ICML 2024) Luca Beurer-Kellner, Marc Fischer, Martin Vechev
  [Code]
• Controlled Text Generation via Language Model Arithmetic (ICLR 2024 – Spotlight) Jasper Dekoninck, Marc Fischer, Luca Beurer-Kellner, Martin Vechev
  [Code]

2023

• LMQL Chat: Scripted Chatbot Development (Neural Conversational AI Workshop – ICML 2023) Luca Beurer-Kellner*, Marc Fischer*, Martin Vechev (*Equal contribution)
  [Code]
• Large Language Models are Zero-Shot Multi-Tool Users (Knowledge and Logical Reasoning Workshop – ICML 2023) Luca Beurer-Kellner*, Marc Fischer*, Martin Vechev (*Equal contribution)
  [Code]
• Prompting Is Programming: A Query Language for Large Language Models (PLDI 2023) Luca Beurer-Kellner, Marc Fischer, Martin Vechev
  [Code]

2022

• Learning to Configure Computer Networks with Neural Algorithmic Reasoning (NeurIPS 2022) Luca Beurer-Kellner, Martin Vechev, Laurent Vanbever, Petar Veličković
  [Code]
• On Distribution Shift in Learning-based Bug Detectors (ICML 2022) Jingxuan He, Luca Beurer-Kellner, Martin Vechev
  [Code]