Hongguang Li (李宏广)
Profile
Head of QifuGPT-Chat Team and Senior Algorithm Expert at AI Lab, Qifu Technology (formerly 360 DigiTech), responsible for Qifu’s full credit-lifecycle Dialogue LLM (collections & telemarketing), and Member of the LLM Special Committee of the Chinese Information Processing Society of China (CIPS). Former Head of LLM Team at JF SmartInvest (9636.HK), leading a team building financial foundation AI agents; Former Applied Scientist of Conversational AI at Microsoft Xiaoice; previously NLP Engineer in Huawei’s Intelligent Automotive BU. Author of 10+ papers (ACL/EMNLP/COLING/NAACL) and holder of 7 granted Chinese invention patents.
Highlights
Key Project Achievements
- Qifu Technology – QifuGPT-Chat: Delivered product-level financial LLM in collections QifuGPT-Chat(0.6B~14B) with SFT and On-Policy Distillation with 10%+ performance improvements(77% ->89%) vs online baseline.
- JF SmartInvest – FinSphere: Built Meta learning RL-based investment advisory agent FinSphere（九方灵犀） in 2025; surpassed major players (e.g., 60% win rate vs East Money agent(东方财富-妙想); 65% win rate vs iWencai（同花顺-问财）).
- Microsoft Xiaoice – COMEDY / XiaobingKnows: Built COMEDY from 0→1 in 2023, a memory-enhanced conversational system, co-developed with HKUST(GZ) and launched on the XEVA platform and Xiaoice Island(小冰岛); Also delivered XiaobingKnows(小冰知道) in 2021, an open-domain QA system using only online search, before ChatGPT’s public release.
- Huawei – Auto Assistant: Core member to the delivery of Huawei’s first-generation in-car assistant（车载小艺） in 2020, deployed across X-Project (AITO/问界), C-Project (AVATR/阿维塔), and B-Project (BAIC/北汽).
Key Research Achievements
- FinSphere & FinSearch: (IJCAI 2025 Workshop Oral, ACM ICAIF 2025): financial agent (co-developed with HKUST Shijie Han) & financial search agent (co-developed with Johns Hopkins Yiqing Shen).
- COMEDY: (COLING 2025 Oral, Top 5%): Open-domain compressive memory-enhanced long-term conversational system, co-developed with HKUST(GZ) Nuo Chen.
- Orca: (EMNLP 2023): First Chinese conversational MRC dataset.
- Penguin: (EMNLP 2023): Largest Chinese generative MRC dataset.
- LAUG: (ACL 2021): Robustness evaluation for task-oriented dialogue, co-developed with Minlie Huang , Chief Scientist of Lingxin Intelligent (AI Beings).
Work Experience
Sep 2025 – Now
Qifu Technology | Senior Expert, Head of QifuGPT-Chat Team |
Advisor:
QingYang, Chief AI Officer
Jul 2024 – Sep 2025
JF Smart Invest(9636.HK) | Senior Manager, Head of LLM Team |
Advisor:
Wenbing Chen, Chairman & CEO
Jul 2021 – Jul 2024
Microsoft Xiaoice | Applied Scientist | Xiaoice |
Advisor:
Baoyuan Wang, Vice President
Jul 2019 – Jul 2021
Huawei | NLP Engineer | Huawei Assistant | Advisor: Yi Gao, Expert
Education
2016 – 2019
University of Science and Technology of China (USTC, 985) | Master of Engineering in Automation |
Advisor:
Jun Li, Associate Professor
2012 – 2016
Donghua University (DHU, 211) | Bachelor of Engineering in Automation |
Advisor:
Wujun Xu, Associate Professor
Publications
Yiqing Shen, Jingshu Zhang, Feng Chen, Kaiyuan Yan, Hongguang Li. FinSearch: A Temporal-Aware Search Agent Framework for Real-Time Financial Information Retrieval with LLMs. ACM ICAIF 2025, Corresponding. DOI
Shijie Han, Jinshu Zhang, YiQing Sheng, Kaiyuan Yang, Hongguang Li. FinSphere, a Real-Time Stock Analysis Agent Powered by Instruction-Tuned LLMs and Domain Tools. IJCAI 2025 Workshop (FinLLM 2025),Oral, Corresponding. DOI
Changhai Zhou, Shijie Han, Lining Yang, Yuhua Zhou, Xu Cheng, Yibin Wang, Hongguang Li. RankAdaptor: Hierarchical Rank Allocation for Efficient Fine-Tuning Pruned LLMs via Performance Model. NAACL (Findings) 2025: 5781-5795, Corresponding. DOI
Changhai Zhou, Yuhua Zhou, Yibin Wang, Shijie Han, Qian Qiao, Hongguang Li. QPruner: Probabilistic Decision Quantization for Structured Pruning in Large Language Models. NAACL (Findings) 2025: 4276-4286, Corresponding. DOI
Nuo Chen, Hongguang Li, Jianhui Chang, Juhua Huang, Baoyuan Wang, Jia Li. Compress to Impress: Unleashing the Potential of Compressive Memory in Real-World Long-Term Conversations. COLING 2025: 755-773,Oral (Top 5%). DOI
Nuo Chen, Hongguang Li, Baoyuan Wang, Jia Li. From Good to Great: Improving Math Reasoning with Tool-Augmented Interleaf Prompting. ACL 2024 Workshop (NLRSE). DOI
Nuo Chen, Hongguang Li, Junqing He, Yinan Bao, Xinshi Lin, Qi Yang, Jianfeng Liu, Ruyi Gan, Jiaxing Zhang, Baoyuan Wang, Jia Li. Orca: A Few-shot Benchmark for Chinese Conversational MRC. EMNLP (Findings) 2023: 15685-15699. DOI
Nuo Chen, Hongguang Li, Yinan Bao, Baoyuan Wang, Jia Li. Natural Response Generation for Chinese Reading Comprehension. EMNLP (Findings) 2023: 11068-11081. DOI
Jiexi Liu, Ryuichi Takanobu, Jiaxin Wen, Dazhen Wan, Hongguang Li, Weiran Nie, Cheng Li, Wei Peng, Minlie Huang. Robustness Testing of Language Understanding in Task-Oriented Dialog. ACL 2021: 2467-2480. DOI
Patents (Authorized)
- CN120030278A (2025): Method and System for Constructing Large Models in the Securities Domain.
- CN118153687A (2024): Memory-Enhanced Response in Dialogue Systems.
- CN117290468B (2024): Intelligent Dialogue Method, Device, and Storage Medium.
- CN115617974A (2023): Dialogue Processing Method, Device, Equipment, and Storage Medium.
- CN115062620A (2022): Semantic Understanding Method and Device.
- CN110134964A (2019): Text Matching Method Based on Hierarchical CNN and Attention.
- CN105749438A (2016): Personal Escape Assistance Device with Embedded Fire Blanket.
Last updated: 2025