Elamur Research⌕
☰
home / papers / 2026 / SeDev: Structured Semantic Exploration for LLM-Driven Code Generation

SeDev: Structured Semantic Exploration for LLM-Driven Code Generation

Paper · 2026 · Ronghui Yang, Jie Liu, Jiajie Zeng et al.

Summary

LLM-based multi-agent systems still struggle to build complete, multi-module software. Feedback-refinement pipelines follow a single trajectory and tend to settle in local optima. MCTS-style search explores more alternatives but becomes very costly. SeDev addresses this by making exploration explicit and cheap.

Starting from a Product Requirement Document, parallel paths each run three agents: Architect, Project Manager and Programmer. Instead of fixed instructions, each agent receives prompts instantiated from a task-agnostic meta-prompt, so instructions become variables to explore. A unit test module scores each attempt. A feedback synthesis module then converts the raw test outcomes into concrete advice, such as switching the programming language, which steers the next round toward semantically nearby solutions.

The authors also release FSD-Bench++, an extension of FSD-bench with richer test cases, and evaluate on it and two public benchmarks. SeDev beats eight baselines after only a few rounds at moderate cost. This suggests that structured, guided exploration can replace expensive tree search for software-level generation.

Key points

  • Results on FSD-Bench++ (GPT-4o-Mini backbone, functional completeness):
    • SeDev reaches 71.96% FC after two rounds and 84.49% after six.
    • The strongest baseline, EvoMAC, scores 62.58%; MapCoder scores 38.42%; GIF-MCTS scores 33.59%.
  • Efficiency: Two SeDev rounds beat EvoMAC and MapCoder, even though both of those run five iterations. Cost in time and tokens rises with each additional round, and this budget is adjustable.
  • Ablation: The test used 15 website tasks. Replacing the meta-prompt with fixed prompts lowered FC from 78.29% to between 71.31% and 76.74%. The largest drop, 6.98 points, came from removing it at the architecture stage.
  • Stated limitations: Test code is generated by LLMs, which may bias evaluation. There is no explicit mechanism for balancing exploration against exploitation. Adding more parallel paths increases the load on the feedback module.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Nanyang Technological UniversitySingapore · Singapore184 works · 27,484 citations🎓Nanjing University of Finance and EconomicsNanjing · China4 works · 2 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksCode & programming languages · 168 works

Cite

Cite as Ronghui Yang, Jie Liu, Jiajie Zeng, Jiexin Wang, Jiuchuan Jiang, Bo An, Yi Cai and Mengchen Zhao. 2026. SeDev: Structured Semantic Exploration for LLM-Driven Code Generation. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1641

@inproceedings{yang2026sedev,
  title = {{SeDev: Structured Semantic Exploration for LLM-Driven Code Generation}},
  author = {Ronghui Yang, Jie Liu, Jiajie Zeng, Jiexin Wang, Jiuchuan Jiang, Bo An, Yi Cai and Mengchen Zhao},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1641},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1641}
}

References

43 references, 20 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Antonis Antoniades, Albert Örwall, Kexun Zhang, Yuxi Xie, Anirudh Goyal, and William Wang. 2024. Swe-search: Enhancing software agents with monte carlo tree search and iterative refinement. arXiv preprint arXiv:2410.20285. find
  2. Youssef Bassil. 2012. A simulation model for the waterfall software development life cycle. ArXiv, abs/1205.6904. — [arXiv:1205.6904](https://arxiv.org/abs/1205.6904) find
  3. Zhangqian Bi, Yao Wan, Zheng Wang, Hongyu Zhang, Batu Guan, Fangxin Lu, Zili Zhang, Yulei Sui, Hai Jin, and Xuanhua Shi. 2024. Iterative refinement of project-level code context for precise code generation with compiler feedback. In Findings of the Association for Computational Linguistics, pages 23… find
  4. Samuel Cahyawijaya, Holy Lovenia, and Pascale Fung. 2024. Llms are few-shot in-context low-resource language learners. arXiv preprint arXiv:2403.16512.
  5. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, and Greg Brockman. 2021a. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374.
  6. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde De Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, and 1 others. 2021b. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374.
  7. Ning Dai, Zheng Wu, Renjie Zheng, Ziyun Wei, Wenlei Shi, Xing Jin, Guanlin Liu, Chen Dun, Liang Huang, and Lin Yan. 2024. Process supervision-guided policy optimization for code generation. arXiv preprint arXiv:2410.17621. find
  8. Nicola Dainese, Matteo Merler, Minttu Alakuijala, and Pekka Marttinen. 2024. Generating code world models with large language models guided by monte carlo tree search. arXiv preprint arXiv:2405.15383. find
  9. Yihong Dong, Xue Jiang, Zhi Jin, and Ge Li. 2024. Self-collaboration code generation via chatgpt. ACM Transactions on Software Engineering and Methodology.
  10. Shihan Dou, Yan Liu, Haoxiang Jia, Enyu Zhou, Limao Xiong, Junjie Shan, Caishuang Huang, Xiao Wang, Xiaoran Fan, Zhiheng Xi, Yuhao Zhou, Tao Ji, Rui Zheng, Qi Zhang, Tao Gui, and Xuanjing Huang. 2024. StepCoder: Improving code generation with reinforcement learning from compiler feedback. In Proceed… find
  11. Daniel Fried, Armen Aghajanyan, Jessy Lin, Sida Wang, Eric Wallace, Freda Shi, Ruiqi Zhong, Scott Yih, Luke Zettlemoyer, and Mike Lewis. 2023. Incoder: A generative model for code infilling and synthesis. In International Conference on Learning Representations. find
  12. Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Jinlin Wang, Ceyao Zhang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, and Jürgen Schmidhuber. 2024. MetaGPT: Meta programming for a multi-agent collaborative framework. In The T…
  13. Yue Hu, Yuzhu Cai, Yaxin Du, Xinyu Zhu, Xiangrui Liu, Zijie Yu, Yuchen Hou, Shuo Tang, and Siheng Chen. 2024. Self-evolving multi-agent collaboration networks for software development. In arXiv preprint arXiv:2410.16946. — [arXiv:2410.16946](https://arxiv.org/abs/2410.16946) find
  14. Dong Huang, Qingwen Bu, Jie M Zhang, Michael Luck, and Heming Cui. 2023. Agentcoder: Multi-agent-based code generation with iterative testing and optimisation. arXiv preprint arXiv:2312.13010.
  15. Md. Ashraful Islam, Mohammed Eunus Ali, and Md Rizwan Parvez. 2024. MapCoder: Multi-agent code generation for competitive problem solving. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, pages 4912– 4944.
  16. Sveidqvist Knut. 2014. Mermaid: Generate diagrams from markdown-like text. find
  17. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. 2022. Large language models are zero-shot reasoners. Advances in neural information processing systems, 35:22199– 22213.
  18. Hung Lea, Yue Wang, Akhilesh Deepak Gotmare, Silvio Savarese, and Steven Chu Hong Hoi. 2022. Coderl: Mastering code generation through pretrained models and deep reinforcement learning. In Advances in Neural Information Processing Systems, pages 21314–21328.
  19. Bowen Li, Wenhan Wu, Ziwei Tang, Lin Shi, John Yang, Jinyang Li, Shunyu Yao, Chen Qian, Binyuan Hui, Qicheng Zhang, Zhiyin Yu, He Du, Ping Yang, Dahua Lin, Chao Peng, and Kai Chen. 2025. Prompting large language models to tackle the full software development lifecycle: A case study. In Proceedings o…
  20. Jiaqi Li, Mengmeng Wang, Zilong Zheng, and Muhan Zhang. 2023. Loogle: Can long-context language models understand long contexts? arXiv preprint arXiv:2311.04939.
  21. Kefan Li and Yuan Yuan. 2024. Large language models as test case generators: Performance evaluation and enhancement. arXiv preprint arXiv:2404.13340. — [arXiv:2404.13340](https://arxiv.org/abs/2404.13340) find
  22. Qingyao Li, Wei Xia, Kounianhua Du, Xinyi Dai, Ruiming Tang, Yasheng Wang, Yong Yu, and Weinan Zhang. 2024. Rethinkmcts: Refining erroneous thoughts in monte carlo tree search for code generation. arXiv preprint arXiv:2409.09584.
  23. Hunter Lightman, Vineet Kosaraju, Yura Burda, Harri Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2023. Let’s verify step by step. arXiv preprint arXiv:2305.20050.
  24. Jie Liu, Guohua Wang, Ronghui Yang, Jiajie Zeng, Mengchen Zhao, and Yi Cai. 2025. RTADev: Intention aligned multi-agent framework for software development. In Findings of the Association for Computational Linguistics: ACL 2025, pages 1548–1581, Vienna, Austria. Association for Computational Linguist… find
  25. Zhe Liu, Chunyang Chen, Junjie Wang, Mengzhuo Chen, Boyu Wu, Xing Che, Dandan Wang, and Qing Wang. 2024. Make llm a testing expert: Bringing human-like interaction to mobile gui testing via functionality-aware decisions. In Proceedings of the IEEE/ACM 46th International Conference on Software Engine… find
  26. Microsoft. 2023. Microsoft attracting users to its code-writing generative ai software. https://www.euronews.com/next/2023/01/25/microsoft-results-ai. ↗
  27. OpenAI. 2022. Chatgpt: Optimizing language models for dialogue. https://openai.com/index/chatgpt/. ↗
  28. Anton Osika. 2023. Gpt-engineer. find
  29. Chen Qian, Xin Cong, Cheng Yang, Weize Chen, Yusheng Su, Juyuan Xu, Zhiyuan Liu, and Maosong Sun. 2023. Communicative agents for software development. arXiv preprint arXiv:2307.07924.
  30. Baptiste Roziere, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, and Jer´ emy Rapin. 2023. Code llama: Open foundation models for code. arXiv preprint arXiv:2308.12950.
  31. Peiyi Wang, Lei Li, Zhihong Shao, Runxin Xu, Damai Dai, Yifei Li, Deli Chen, Yu Wu, and Zhifang Sui. 2024a. Math-shepherd: Verify and reinforce llms step-by-step without human annotations. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pape…
  32. Xin Wang, Yasheng Wang, Yao Wan, Fei Mi, Yitong Li, Pingyi Zhou, Jin Liu, Hao Wu, Xin Jiang, and Qun Liu. 2022. Compilable neural code generation with compiler feedback. In Findings of the Association for Computational Linguistics, pages 9–19. find
  33. Xindi Wang, Mahsa Salmani, Parsa Omidi, Xiangyu Ren, Mehdi Rezagholizadeh, and Armaghan Eshaghi. 2024b. Beyond the limits: A survey of techniques to extend the context length in large language models. arXiv preprint arXiv:2402.02244.
  34. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824– 24837.
  35. Sunny Wong, Yuanfang Cai, Miryung Kim, and Michael Dalton. 2011. Detecting software modularity violations. In Proceedings of the 33rd International Conference on Software Engineering, pages 411–420. find
  36. Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Shaokun Zhang, Erkang Zhu, Beibin Li, Li Jiang, Xiaoyun Zhang, and Chi Wang. 2023. Autogen: Enabling next-gen llm applications via multi-agent conversation framework. arXiv preprint arXiv:2308.08155.
  37. Bin Xu, Yiguan Lin, Yinghao Li, and 1 others. 2024. Sra-mcts: Self-driven reasoning aurmentation with monte carlo tree search for enhanced code generation. arXiv preprint arXiv:2411.11053. — [arXiv:2411.11053](https://arxiv.org/abs/2411.11053) find
  38. Li Yuan, Yi Cai, Xudong Shen, Qing Li, Qingbao Huang, Zikun Deng, and Tao Wang. 2025a. Collaborative multi-lora experts with achievement-based multi-tasks loss for unified multimodal information extraction. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence… find
  39. Li Yuan, Qingfei Huang, Bingshan Zhu, Yi Cai, Qingbao Huang, Changmeng Zheng, Zikun Deng, and Tao Wang. 2026. Hybrid-dmkg: A hybrid reasoning framework over dynamic multimodal knowledge graphs for multimodal multihop qa with knowledge editing. In Proceedings of the AAAI Conference on Artificial Inte… find
  40. Shuozhi Yuan, Liming Chen, Miaomiao Yuan, Jin Zhao, Haoran Peng, and Wenming Guo. 2025b. Mcts-sql: An effective framework for text-to-sql with monte carlo tree search. arXiv preprint arXiv:2501.16607. — [arXiv:2501.16607](https://arxiv.org/abs/2501.16607) find
  41. Sai Zhang, Zhenchang Xing, Ronghui Guo, Fangzhou Xu, Lei Chen, Zhaoyuan Zhang, Xiaowang Zhang, Zhiyong Feng, and Zhiqiang Zhuang. 2025. Empowering agile-based generative software development through human-ai teamwork. 34(6). find
  42. Shun Zhang, Zhenfang Chen, Yikang Shen, Mingyu Ding, Joshua B Tenenbaum, and Chuang Gan. 2023. find
  43. Planning with large language models for code generation. In International Conference on Learning Representations.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2026-sedev-structured-semantic-exploration-llm/