@inproceedings{shbita2025mermaidseqbench,
  title={MermaidSeqBench: An Evaluation Benchmark for LLM-to-Mermaid Sequence Diagram Generation},
  author={Shbita, Basel and Ahmed, Farhan and DeLuca, Chad},
  booktitle={NeurIPS 2025 Workshop on Evaluating the Evolving LLM Lifecycle: Benchmarks, Emergent Abilities, and Scaling},
  year={2025},
  doi={10.48550/arXiv.2511.14967},
  urlLink={https://arxiv.org/abs/2511.14967},
  urlPaper={https://arxiv.org/pdf/2511.14967.pdf}
}

{"_id":"oFtm26pBzHWWCnTYM","bibbaseid":"shbita-ahmed-deluca-mermaidseqbenchanevaluationbenchmarkforllmtomermaidsequencediagramgeneration-2025","author_short":["Shbita, B.","Ahmed, F.","DeLuca, C."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","title":"MermaidSeqBench: An Evaluation Benchmark for LLM-to-Mermaid Sequence Diagram Generation","author":[{"propositions":[],"lastnames":["Shbita"],"firstnames":["Basel"],"suffixes":[]},{"propositions":[],"lastnames":["Ahmed"],"firstnames":["Farhan"],"suffixes":[]},{"propositions":[],"lastnames":["DeLuca"],"firstnames":["Chad"],"suffixes":[]}],"booktitle":"NeurIPS 2025 Workshop on Evaluating the Evolving LLM Lifecycle: Benchmarks, Emergent Abilities, and Scaling","year":"2025","doi":"10.48550/arXiv.2511.14967","urllink":"https://arxiv.org/abs/2511.14967","urlpaper":"https://arxiv.org/pdf/2511.14967.pdf","bibtex":"@inproceedings{shbita2025mermaidseqbench,\n title={MermaidSeqBench: An Evaluation Benchmark for LLM-to-Mermaid Sequence Diagram Generation},\n author={Shbita, Basel and Ahmed, Farhan and DeLuca, Chad},\n booktitle={NeurIPS 2025 Workshop on Evaluating the Evolving LLM Lifecycle: Benchmarks, Emergent Abilities, and Scaling},\n year={2025},\n doi={10.48550/arXiv.2511.14967},\n urlLink={https://arxiv.org/abs/2511.14967},\n urlPaper={https://arxiv.org/pdf/2511.14967.pdf}\n}\n\n\n","author_short":["Shbita, B.","Ahmed, F.","DeLuca, C."],"key":"shbita2025mermaidseqbench","id":"shbita2025mermaidseqbench","bibbaseid":"shbita-ahmed-deluca-mermaidseqbenchanevaluationbenchmarkforllmtomermaidsequencediagramgeneration-2025","role":"author","urls":{"Link":"https://arxiv.org/abs/2511.14967","Paper":"https://arxiv.org/pdf/2511.14967.pdf"},"metadata":{"authorlinks":{}},"downloads":2},"bibtype":"inproceedings","biburl":"https://shbita.com/docs/shbita.bib","dataSources":["8eQQPu5qWjk8KSph6","Qx7EhmagS9HmZqzQr"],"keywords":[],"search_terms":["mermaidseqbench","evaluation","benchmark","llm","mermaid","sequence","diagram","generation","shbita","ahmed","deluca"],"title":"MermaidSeqBench: An Evaluation Benchmark for LLM-to-Mermaid Sequence Diagram Generation","year":2025,"downloads":2}