I work on web agents and text embeddings, with a focus on building benchmarks and tools to evaluate them.
Web agents
- WebArena-Pro: a heterogeneous, multimodal and reproducible benchmark for web agents. NeurIPS E&D Track, 2026 · 📄 Paper · 💻 Code
- FocusAgent: trims the large context of web agents with a small retriever that keeps only the lines of the page that matter, which also filters out prompt injections. 📄 Paper · 💻 Code
- CUBE: a standard for unifying agent benchmarks. 📄 Paper · 💻 Code
- Planning representations for web agents: an empirical study of how the way LLM web agents plan affects their performance. 📄 Paper
Embedding benchmarks
- MTEB: the Massive Text Embedding Benchmark, which I contribute to.
- MTEB-French: resources for evaluating French sentence embeddings. EGC 2025 · 📄 Paper
- Web agents and how to evaluate them
- Text embeddings and benchmarking




