A personal homepage about multi-agent systems and applied large language models — how agents collaborate, how they catch each other's mistakes, and how research becomes something usable.
Same-model multi-role and multi-model heterogeneous architectures, debate and consensus mechanisms, failure modes and recovery — the main focus here.
Putting model capability into real tasks: evaluation, tool use, long-horizon work, and reproducible experiments.
Tracking new papers on arXiv and Hugging Face — and keeping the judgment after reading, not just the title.