Lev Sorokin

PhD Researcher – Generative AI at BMW Group • Automated Software Testing group at TUM

prof_pic.jpg

BMW Group

Munich, Germany

I am a PhD researcher in Generative AI at the BMW Group and part of the Automated Software Testing group of Andrea Stocco at the Technical University of Munich (TUM). Before that, I was a research and development engineer at fortiss, working on testing and assurance of autonomous systems.

My research focuses on making AI-enabled systems reliable and safe through automated testing.

Current research themes:

  • Testing LLM-based applications :speech_balloon: — search-based testing of LLM apps, vision-language models, and in-car conversational assistants [ICSE 2027, SANER 2026, ASE 2026, JSS 2026, ITSC 2026]
  • Simulation-based testing of automated driving systems :oncoming_automobile: — multi-fidelity testing and simulator ensembles [ICSE 2027, EMSE 2026]

news

Oct 21, 2026 :pushpin: The paper SimFusion: Multi-Fidelity Simulation-based Testing of Automated Driving Systems has been accepted for publication at ICSE 2027!
Jul 14, 2026 :microphone: I gave an invited talk at the TAROT Summer School 2026 in Munich on testing, validation, and security of AI-enabled systems!
Jul 01, 2026 :pushpin: The paper Search-based Testing of Vision Language Models for In-Car Scene Understanding has been accepted for publication at ASE 2026 - Industry Showcase Track!
May 01, 2026 :pushpin: The paper Automated Evaluation of Multi-Turn Dialogues in In-Car Conversational Assistants has been accepted for publication at ITSC 2026!
Apr 21, 2026 :pushpin: The paper Benchmarking Contextual Understanding for In-Car Conversational Systems has been accepted for publication in The Journal of Systems & Software!
Mar 20, 2026 :trophy: The paper STELLAR: A Search-Based Testing Framework for Large Language Model Applications received the IEEE Computer Society TCSE Distinguished Paper Award at SANER 2026!
Feb 02, 2026 :pushpin: The paper Simulator Ensembles for Trustworthy Autonomous Driving Testing has been accepted for publication in Empirical Software Engineering!
Dec 09, 2025 :pushpin: The paper STELLAR: A Search-Based Testing Framework for Large Language Model Applications has been accepted for publication at SANER 2026!

highlights

STELLAR icon
STELLAR SANER'26 Distinguished Paper

Search-based testing framework for LLM-based applications.

OpenSBT icon
OpenSBT

Modular framework for search-based testing of automated driving systems.

SimFusion icon
SimFusion

Multi-fidelity simulation-based testing of automated driving systems.

MultiSim icon
MultiSim

Simulator ensembles for trustworthy testing of autonomous driving systems.

ISU-Test icon
ISU-Test

Search-based testing of vision-language models for in-car scene understanding.

CyberDS icon
CyberDS

Auditable runtime monitoring of distributed cloud systems with Datalog-based specifications and tamper-proof logs.