Elamur Research⌕
☰
home / papers / 2026 / How to Train a Real-World Silicon Concierge? Internalizing Complex Business Workflow to Only OneModel

How to Train a Real-World Silicon Concierge? Internalizing Complex Business Workflow to Only OneModel

Paper · 2026 · Yongqi Tong, Xiaoyun Feng, Lyuxin Xue et al.

Summary

Industrial customer-service agents are usually built as chains of separate modules: routing, retrieval, planning, execution, response writing and review, as in the PEER pipeline. The authors argue that these chains pile up patches, pass errors downstream, are hard to debug and slow to respond. Their alternative, OneModel, moves the business workflow into the weights of a single compact model.

Knowledge is sorted by how often it changes:

  • Stable domain principles are taught through Continual Pre-training (CPT). This includes a reflection step in which small SFT probes guide the synthesis of the CPT corpus.
  • Semi-stable procedures (SOPs, reply templates, basic tool use) are trained in through SFT with self-reflection rationales.
  • Remaining weak spots are handled with multi-round RL using simulated users and multi-objective rewards.
  • Fast-changing data, such as exchange rates, is supplied in the prompt at runtime.

In weeks of online A/B tests on Ant International's Global Merchant Service Agent, OneModel roughly halved latency and clearly raised the share of queries resolved without humans. This suggests unified models can replace brittle workflow engineering in high-stakes finance.

Key points

  • Deployment: the system runs on Ant International's Global Merchant Service Agent. A single 8B OneModel stands in for six or more specialized PEER components.
  • Online A/B results:
    • End-to-end latency fell from 18.7s to 8s, a 57.2% reduction.
    • Intelligent Resolution Rate (IRR) rose from 64.3% to 83.3%.
  • Comparison with commercial models: OneModel's peak resolution rate was 90.75%, versus 86.72% for Claude-3.5-Haiku and 87.55% for both Gemini-2.5-Pro and GPT-5.2.
  • User simulator problem: the simulator's "satisfaction" signal barely matched real user behavior (Kappa 0.0446). Simulated users asked for a human in 1.7% of cases, against about 10% for real users.
  • Other lessons:
    • A reward model trained on cleaned annotation data beat larger models trained on noisy labels.
    • Putting volatile rules into CPT led to "toxic knowledge" that could not be updated safely.
    • RL tended to make the model sycophantic, so compliance violations were made a hard veto in the reward.

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

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksEngineering · 133 worksIndustrial and Manufacturing Engineering · 10 works

Cite

Cite as Yongqi Tong, Xiaoyun Feng, Lyuxin Xue, Jianshe Li, Xin Zhang and Jiang-Ming Yang. 2026. How to Train a Real-World Silicon Concierge? Internalizing Complex Business Workflow to Only OneModel. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.91

@inproceedings{tong2026how,
  title = {{How to Train a Real-World Silicon Concierge? Internalizing Complex Business Workflow to Only OneModel}},
  author = {Yongqi Tong, Xiaoyun Feng, Lyuxin Xue, Jianshe Li, Xin Zhang and Jiang-Ming Yang},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.91},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.91}
}

References

42 references, 22 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Victor Barres, Honghua Dong, Soham Ray, Xujie Si, and Karthik Narasimhan. 2025. τ 2 -bench: Evaluating conversational agents in a dual-control environment. arXiv preprint arXiv:2506.07982. find
  2. Mengyuan Chen, Chengjun Dai, Xinyang Dong, Chengzhe Feng, Kewei Fu, Jianshe Li, Zhihan Peng, Yongqi Tong, Junshao Zhang, and Hong Zhu. 2025. find
  3. Dingtalk deepresearch: A unified multi agent framework for adaptive intelligence in enterprise environments. Preprint, arXiv:2510.24760. — [arXiv:2510.24760](https://arxiv.org/abs/2510.24760) find
  4. Tianzhe Chu, Yuexiang Zhai, Jihan Yang, Shengbang Tong, Saining Xie, Dale Schuurmans, Quoc V. Le, Sergey Levine, and Yi Ma. 2025. Sft memorizes, rl generalizes: A comparative study of foundation model post-training. Preprint, arXiv:2501.17161.
  5. DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, and 181 others. 2025. Deepseek-r1: Incentivizing reasoning capability in ll…
  6. Ning Ding, Yulin Chen, Bokai Xu, Yujia Qin, Shengding Hu, Zhiyuan Liu, Maosong Sun, and Bowen Zhou. 2023. Enhancing chat language models by scaling high-quality instructional conversations. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 3029–3051.
  7. Nouha Dziri, Ximing Lu, Melanie Sclar, Xiang Lorraine Li, Liwei Jiang, Bill Yuchen Lin, Peter West, Chandra Bhagavatula, Ronan Le Bras, Jena D. Hwang, Soumya Sanyal, Sean Welleck, Xiang Ren, Allyson Ettinger, Zaid Harchaoui, and Yejin Choi. 2023. Faith and fate: Limits of transformers on composition…
  8. Suchin Gururangan, Ana Marasović, Swabha Swayamdipta, Kyle Lo, Iz Beltagy, Doug Downey, and Noah A. Smith. 2020. Don’t stop pretraining: Adapt language models to domains and tasks. Preprint, arXiv:2004.10964. find
  9. Haoyang He, Zihua Rong, Kun Ji, Chenyang Li, Qing Huang, Chong Xia, Lan Yang, and Honggang Zhang. 2025. Rethinking reasoning quality in large language models through enhanced chain-of-thought via rl. Preprint, arXiv:2509.06024. — [arXiv:2509.06024](https://arxiv.org/abs/2509.06024) find
  10. Sirui Hong, Mingchen Zhuge, Jiaqi Chen, Xiawu Zheng, Yuheng Cheng, Ceyao Zhang, Jinlin Wang, 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. Preprint, a…
  11. Jie Huang, Xinyun Chen, Swaroop Mishra, Huaixiu Steven Zheng, Adams Wei Yu, Xinying Song, and Denny Zhou. 2024. Large language models cannot self-correct reasoning yet. Preprint, arXiv:2310.01798.
  12. Nikhil Kandpal, Haikang Deng, Adam Roberts, Eric Wallace, and Colin Raffel. 2023. Large language models struggle to learn long-tail knowledge. Preprint, arXiv:2211.08411.
  13. Omar Khattab, Arnav Singhvi, Paridhi Maheshwari, Zhiyuan Zhang, Keshav Santhanam, Sri Vardhamanan, Saiful Haq, Ashutosh Sharma, Thomas T. Joshi, Hanna Moazam, Heather Miller, Matei Zaharia, and Christopher Potts. 2023. Dspy: Compiling declarative language model calls into self-improving pipelines. P…
  14. Chuyi Kong, Yaxin Fan, Xiang Wan, Feng Jiang, and Benyou Wang. 2024. Platolm: Teaching llms in multi-round dialogue via a user simulator. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7841–7863. find
  15. Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, and Ion Stoica. 2023. Efficient memory management for large language model serving with pagedattention. Preprint, arXiv:2309.06180.
  16. Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, and 1 others. 2024. Tulu 3: Pushing frontiers in open language model post-training. arXiv preprint arXiv:2411.15124.
  17. Angeliki Lazaridou, Adhiguna Kuncoro, Elena Gribovskaya, Devang Agrawal, Adam Liska, Tayfun Terzi, Mai Gimenez, Cyprien de Masson d’Autume, Tomas Kocisky, Sebastian Ruder, Dani Yogatama, Kris Cao, Susannah Young, and Phil Blunsom. 2021. Mind the gap: Assessing temporal generalization in neural langu… find
  18. 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. Preprint, arXiv:2305.20050.
  19. Hsien-chin Lin, Nurul Lubis, Songbo Hu, Carel van Niekerk, Christian Geishauser, Michael Heck, Shutong Feng, and Milica Gasic. 2021. Domain-independent user simulation with transformers for task-oriented dialogue systems. In Proceedings of the 22nd Annual Meeting of the Special Interest Group on Dis… find
  20. Zi Lin, Zihan Wang, Yongqi Tong, Yangkun Wang, Yuxin Guo, Yujia Wang, and Jingbo Shang. 2023. Toxicchat: Unveiling hidden challenges of toxicity detection in real-world user-ai conversation. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 4694–4702.
  21. Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2023. Lost in the middle: How language models use long contexts. Preprint, arXiv:2307.03172.
  22. Xiang Luo, Zhiwen Tang, Jin Wang, and Xuejie Zhang. 2024. Duetsim: Building user simulator with dual large language models for task-oriented dialogues. arXiv preprint arXiv:2405.13028. new framework to guide llms’ non-linear thinking. Preprint, arXiv:2310.12342. find
  23. Licheng Pan, Yongqi Tong, Xin Zhang, Xiaolu Zhang, Jun Zhou, and Zhixuan Chu. 2025. Understanding and mitigating overrefusal in LLMs from an unveiling perspective of safety decision boundary. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 21057–21075… find
  24. Jonathan Uesato, Nate Kushman, Ramana Kumar, Francis Song, Noah Siegel, Lisa Wang, Antonia Creswell, Geoffrey Irving, and Irina Higgins. 2022. Solving math word problems with process- and outcome-based feedback. Preprint, arXiv:2211.14275.
  25. Ofir Press, Muru Zhang, Sewon Min, Ludwig Schmidt, Noah A. Smith, and Mike Lewis. 2023. Measuring and narrowing the compositionality gap in language models. Preprint, arXiv:2210.03350. find
  26. Jost Schatzmann, Blaise Thomson, Karl Weilhammer, Hui Ye, and Steve Young. 2007. Agenda-based user simulation for bootstrapping a pomdp dialogue system. In Human Language Technologies 2007: The Conference of the North American Chapter of the Association for Computational Linguistics; Companion Volum…
  27. Ivan Sekulić, Silvia Terragni, Victor Guimarães, Nghia Khau, Bruna Guedes, Modestas Filipavicius, Andre Ferreira Manso, and Roland Mathis. 2024. Reliable llm-based user simulator for task-oriented dialogue systems. arXiv preprint arXiv:2402.13374. find
  28. Yunfan Shao, Linyang Li, Junqi Dai, and Xipeng Qiu. 2023. Character-llm: A trainable agent for role-playing. arXiv preprint arXiv:2310.10158.
  29. Yuchong Sun, Che Liu, Kun Zhou, Jinwen Huang, Ruihua Song, Wayne Xin Zhao, Fuzheng Zhang, Di Zhang, and Kun Gai. 2024. Parrot: Enhancing multi-turn instruction following for large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: L… find
  30. Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, Chuning Tang, Congcong Wang, Dehao Zhang, Enming Yuan, Enzhe Lu, Fengxiang Tang, Flood Sung, Guangda Wei, Guokun Lai, and 77 others. 2025. Kimi k1.5: Scaling reinforcement l…
  31. Yongqi Tong, Dawei Li, Sizhe Wang, Yujia Wang, Fei Teng, and Jingbo Shang. 2024a. Can llms learn from previous mistakes? investigating llms’ errors to boost for reasoning. Preprint, arXiv:2403.20046.
  32. Yongqi Tong, Sizhe Wang, Dawei Li, Yifan Wang, Simeng Han, Zi Lin, Chengsong Huang, Jiaxin Huang, and Jingbo Shang. 2024b. Optimizing language model’s reasoning abilities with weak supervision. Preprint, arXiv:2405.04086. — [arXiv:2405.04086](https://arxiv.org/abs/2405.04086) find
  33. Kuang Wang, Xianfei Li, Shenghao Yang, Li Zhou, Feng Jiang, and Haizhou Li. 2025a. Know you first and be you better: Modeling human-like user simulators via implicit profiles. arXiv preprint arXiv:2502.18968. find
  34. Lei Wang, Wanyu Xu, Yihuai Lan, Zhiqiang Hu, Yunshi Lan, Roy Ka-Wei Lee, and Ee-Peng Lim. 2023. Plan-and-solve prompting: Improving zero-shot chain-of-thought reasoning by large language models. Preprint, arXiv:2305.04091. find
  35. Sizhe Wang, Yongqi Tong, Hengyuan Zhang, Dawei Li, Xin Zhang, and Tianlong Chen. 2025b. Bpo: Towards balanced preference optimization between knowledge breadth and depth in alignment. Preprint, arXiv:2411.10914. find
  36. Yiying Wang, Xiaojing Li, Binzhu Wang, Yueyang Zhou, Yingru Lin, Han Ji, Hong Chen, Jinshi Zhang, Fei Yu, Zewei Zhao, Song Jin, Renji Gong, and Wanqing Xu. 2024. Peer: Expertizing domain-specific tasks with a multi-agent framework and tuning methods. Preprint, arXiv:2407.06985. — [arXiv:2407.06985](https://arxiv.org/abs/2407.06985) find
  37. Canwen Xu, Daya Guo, Nan Duan, and Julian McAuley. 2023. Baize: An open-source chat model with parameter-efficient tuning on self-chat data. arXiv preprint arXiv:2304.01196.
  38. Zhihao Xu, Yongqi Tong, Xin Zhang, Jun Zhou, and Xiting Wang. 2025. Reward consistency: Improving multi-objective alignment from a data-centric perspective. Preprint, arXiv:2504.11337. — [arXiv:2504.11337](https://arxiv.org/abs/2504.11337) find
  39. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. Preprint, arXiv:2210.03629.
  40. Edward Yeo, Yuxuan Tong, Morry Niu, Graham Neubig, and Xiang Yue. 2025. Demystifying long chain-of-thought reasoning in llms. Preprint, arXiv:2502.03373.
  41. Ori Yoran, Tomer Wolfson, Ori Ram, and Jonathan Berant. 2024. Making retrieval-augmented language models robust to irrelevant context. Preprint, arXiv:2310.01558.
  42. Yongqi Tong, Yifan Wang, Dawei Li, Sizhe Wang, Zi Lin, Simeng Han, and Jingbo Shang. 2023. Eliminating reasoning via inferring with planning: A find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tong-2026-how-train-real-world-silicon/