Hi — curious minds of the internet. I am Saber Zerhoudi, currently an Information Retrieval research scientist, based in Passau, Germany. My current research focuses on user simulation, user-centric evaluation of interactive search systems, and agentic AI.

Previously, I earned my Ph.D. at University of Passau, where I contributed to the OpenWebSearch project on large-scale, distributed web crawling and the creation of an open web index, and to the SINIR project on simulation-based evaluation of interactive search behavior in digital libraries.

I'm a user simulation and search researcher. I'm broadly interested in how we can evaluate and design information access systems from a user-centric perspective, especially now that generative models and agents are part of the picture. My work falls into three interconnected themes:

• User Simulation: I build platforms and benchmarks for simulating realistic search behavior across classic, conversational, and agentic settings. My recent work centers on the IIRSim ecosystem, which includes SimEval-IR, a unified toolkit and benchmark suite that disentangles behavioral realism from tester reliability when evaluating simulators; IIRSim Studio, a dashboard for configuring and running simulation studies; and UXSim, a hybrid simulator that fuses cognitive layers with LLM-driven personas. Together, these resources provide a reproducible testbed for systematic experimentation across diverse user types, interfaces, and tasks.

• Open Web Crawling & Indexing: I help design and operate large-scale crawling pipelines and infrastructure for the European Open Web Index, this involves the Open Web Search Crawler (OWLer) and its ecosystem for distributed, efficient crawling. Furthermore, I contribute to services exposing the Open Web Index for search, analytics, and generative AI applications.

• User-Centric RAG & Agents: I study how retrieval-augmented generation and agentic systems can stay trustworthy, governable, and user-aware as content evolves and as the "user" itself becomes an LLM-driven agent. A central recent contribution is NuggetIndex, which stores atomic information units as governed records with evidence links, temporal validity, and lifecycle state, preventing outdated or contested facts from reaching the generator. Complementing this, AgentSim produces verifiable, stepwise reasoning traces over user-defined corpora, and PersonaRAG uses cooperating agents to maintain stable user personas that guide retrieval and generation. Together, these systems push RAG beyond passage retrieval toward maintainable, accountable, and user-aware information access.

Additionally, I founded SearchSim, a research initiative studying how information access changes when autonomous agents search, decide, and act alongside humans.

At the University of Passau, I am involved in teaching and supporting courses in: