Fengyu Cai
Ph.D. Student at TU Darmstadt
{first name}.{last name} [AT] tu-darmstadt.de
Bio
I am a final-year Ph.D. student at TU Darmstadt, Germany, co-supervised by Heinz Koeppl and Iryna Gurevych. I expect to graduate by the end of 2026 and am on the job market.
My research lies at the intersection of Large Language Models, Information Retrieval, RAG, and Reinforcement Learning. I am particularly interested in (1) training retrievers with language modeling rather than contrastive pairs, so that strong retrievers can be built cheaply in specialized domains such as science and code (Revela, ElicitR, CoQuIR); (2) retrieval granularity and embedding geometry (MixGR, GeoHard); and (3) agentic RL for LLM post-training on long-horizon tasks.
In summer 2026 I was a Research Intern at Microsoft (Office of Applied Research, Redmond), working on long-horizon agentic RL for LLM post-training. Before my Ph.D., I obtained my M.Sc. in Computer Science at EPFL, supervised by Boi Faltings, and my B.Eng. in Electrical Engineering and B.Sc. in Computer Science at HKUST.
Highlights
- 🏆 Best Paper Award, FrontierIR@AAAI 2026 (Revela, first author)
- 🏆 SAC Highlight, ACL 2026 — top 0.5% of submissions (CoQuIR, co-first author)
- 🎤 Oral presentations at ICLR 2026 (top 1.1%, Revela) and ACL 2026 (CoQuIR)
- 💶 Awardee of a BMBF Software Campus grant (€115,000) on trustworthy RAG in industrial domains
- 🤗 Revela models and data: 7K+ downloads on Hugging Face
- 🧊 Contributor to the LCO-Embedding open-source omnimodal embedding model series (text/image/audio/video), 130K+ downloads on Hugging Face across the series; core contributor to the latest LCO-Embedding-Omni-3B-2605 release
News
- 2026.10 — Invited talk at the AI TIME Emerging Researchers Speaker Program: Revela: Dense Retriever Learning via Language Modeling.
- 2026.09 — Crayotter accepted to AACL-IJCNLP 2026 System Demonstrations.
- 2026.07 — Invited talk at the AgentSearch Workshop @ SIGIR 2026, Melbourne: Beyond Contrastive Pairs: Language Modeling as a Learning Signal for Dense Retrieval.
- 2026.07 — CoQuIR featured by 青稞AI.
- 2026.06 — Started as a Research Intern at Microsoft, Redmond (Office of Applied Research).
- 2026.05 — 🏆 CoQuIR accepted to ACL 2026 as an Oral and selected as SAC Highlight (top 0.5% of submissions).
- 2026.05 — ElicitR and Epistemic Gain, Aleatoric Cost accepted to ICML 2026; recognized as a Gold Reviewer at ICML 2026.
- 2026.03 — Revela featured by 机器之心 (Synced).
- 2026.02 — CodeMMR accepted to CVPR 2026.
- 2026.01 — 🏆 Revela accepted to ICLR 2026 as an Oral (top 1.1%) and won the Best Paper Award at FrontierIR@AAAI 2026. Models and data are on Hugging Face (7K+ downloads).
- 2025 — Awarded a BMBF Software Campus grant (€115,000) and Hessian AI computation grants (235,000 GPU hours).
Publications
Most recent publications on Google Scholar.
‡ indicates equal contribution.
CoQuIR: A Comprehensive Benchmark for Code Quality-Aware Information Retrieval
Jiahui Geng‡, Fengyu Cai‡, Shaobo Cui, Qing Li, Liangwei Chen, Chenyang Lyu, Haonan Li, Derui Zhu, Walter Pretschner, Heinz Koeppl, Fakhri Karray
ACL 2026 Main Conference — 🎤 Oral, 🏆 SAC Highlight (top 0.5% of submissions)
ElicitR: Unlocking Latent Reasoning in Dense Retrievers via Generative Regularization
Fengyu Cai, Iryna Gurevych, Heinz Koeppl
ICML 2026
Revela: Dense Retriever Learning via Language Modeling
Fengyu Cai, Tong Chen, Xinran Zhao, Sihao Chen, Hongming Zhang, Sherry Tongshuang Wu, Iryna Gurevych, Heinz Koeppl
ICLR 2026 — 🎤 Oral (top 1.1%); 🏆 Best Paper Award at FrontierIR@AAAI 2026
Knowledge Graph–Augmented DNA Representation Learning
Fengyu Cai‡, Erik Kubaczka‡, Shaobo Cui, Heinz Koeppl
ICML 2025 Workshop on Multi-modal Foundation Models and LLMs for Life Sciences
MoR: Better Handling Diverse Queries with a Mixture of Sparse, Dense, and Human Retrievers
Jushaan Singh Kalra, Xinran Zhao, To Eun Kim, Fengyu Cai, Fernando Diaz, Tongshuang Wu
EMNLP 2025 Main Conference
MixGR: Enhancing Retriever Generalization for Scientific Domain through Complementary Granularity
Fengyu Cai, Xinran Zhao, Tong Chen, Sihao Chen, Hongming Zhang, Iryna Gurevych, Heinz Koeppl
EMNLP 2024 Main Conference
GeoHard: Towards Measuring Class-wise Hardness through Modelling Class Semantics
Fengyu Cai, Xinran Zhao, Hongming Zhang, Iryna Gurevych, Heinz Koeppl
ACL 2024 Findings
A Survey of Confidence Estimation and Calibration in Large Language Models
Jiahui Geng, Fengyu Cai, Yuxia Wang, Heinz Koeppl, Preslav Nakov, Iryna Gurevych
NAACL 2024 Main Conference
CoQuIR: A Comprehensive Benchmark for Code Quality-Aware Information Retrieval
Jiahui Geng‡, Fengyu Cai‡, Shaobo Cui, Qing Li, Liangwei Chen, Chenyang Lyu, Haonan Li, Derui Zhu, Walter Pretschner, Heinz Koeppl, Fakhri Karray
ACL 2026 Main Conference — 🎤 Oral, 🏆 SAC Highlight (top 0.5% of submissions)
ElicitR: Unlocking Latent Reasoning in Dense Retrievers via Generative Regularization
Fengyu Cai, Iryna Gurevych, Heinz Koeppl
ICML 2026
Epistemic Gain, Aleatoric Cost: Uncertainty Decomposition in Multi-Agent Debate for Math Reasoning
Dan Qiao, Binbin Chen, Fengyu Cai, Jianlong Chen, Wenhao Li, Fuxin Jiang, Zuzhi Chen, Hongyuan Zha, Tieying Zhang, Baoxiang Wang
ICML 2026
CodeMMR: Bridging Natural Language, Code, and Image for Unified Retrieval
Jiahui Geng, Qing Li, Fengyu Cai, Fakhri Karray
CVPR 2026
Revela: Dense Retriever Learning via Language Modeling
Fengyu Cai, Tong Chen, Xinran Zhao, Sihao Chen, Hongming Zhang, Sherry Tongshuang Wu, Iryna Gurevych, Heinz Koeppl
ICLR 2026 — 🎤 Oral (top 1.1%); 🏆 Best Paper Award at FrontierIR@AAAI 2026
Crayotter: Traceable Multi-Agent Workflows for Long-Form Video Editing
Lecheng Yan, Yichong Zhang, Xiantao Xu, Jianze Lin, Ben Pan, Xiaoyu Zheng, Jiawei Qian, Anqi Wu, Jiahui Geng, Ruizhe Li, Fengyu Cai, Jingcheng Niu, Raymond Li, Wenxi Li, Chenyang Lyu
AACL-IJCNLP 2026 System Demonstrations
Knowledge Graph–Augmented DNA Representation Learning
Fengyu Cai‡, Erik Kubaczka‡, Shaobo Cui, Heinz Koeppl
ICML 2025 Workshop on Multi-modal Foundation Models and LLMs for Life Sciences
MoR: Better Handling Diverse Queries with a Mixture of Sparse, Dense, and Human Retrievers
Jushaan Singh Kalra, Xinran Zhao, To Eun Kim, Fengyu Cai, Fernando Diaz, Tongshuang Wu
EMNLP 2025 Main Conference
A Survey of Machine Unlearning in Large Language Models: Methods, Challenges and Future Directions
Qing Li, Jiahui Geng, Herbert Woisetschläger, Zongxiong Chen, Fengyu Cai, Yuxia Wang, Preslav Nakov, Hans-Arno Jacobsen, Fakhri Karray
arXiv preprint, 2025
MixGR: Enhancing Retriever Generalization for Scientific Domain through Complementary Granularity
Fengyu Cai, Xinran Zhao, Tong Chen, Sihao Chen, Hongming Zhang, Iryna Gurevych, Heinz Koeppl
EMNLP 2024 Main Conference
Aligning Large Language Model with Direct Multi-Preference Optimization for Recommendation
Zhuoxi Bai, Ning Wu, Fengyu Cai, Xinyi Zhu, Xiangnan He
CIKM 2024
GeoHard: Towards Measuring Class-wise Hardness through Modelling Class Semantics
Fengyu Cai, Xinran Zhao, Hongming Zhang, Iryna Gurevych, Heinz Koeppl
ACL 2024 Findings
A Survey of Confidence Estimation and Calibration in Large Language Models
Jiahui Geng, Fengyu Cai, Yuxia Wang, Heinz Koeppl, Preslav Nakov, Iryna Gurevych
NAACL 2024 Main Conference
SLIM: Explicit Slot-Intent Mapping with BERT for Joint Multi-Intent Detection and Slot Filling
Fengyu Cai, Wanhao Zhou, Fei Mi, Boi Faltings
ICASSP 2022
Self-training Improves Pre-training for Few-shot Learning in Task-oriented Dialogue Systems
Fei Mi, Wanhao Zhou, Fengyu Cai, Lingjing Kong, Minlie Huang, Boi Faltings
EMNLP 2021 Main Conference
A Collaborative Aerial-Ground Robotic System for Fast Exploration
Luqi Wang, Daqian Cheng, Fei Gao, Fengyu Cai, Jixin Guo, Mengxiang Lin, Shaojie Shen
ISER 2018 (Proceedings of the International Symposium on Experimental Robotics)
CRASH: A Collaborative Aerial-Ground Exploration System Using Hybrid-Frontier Method
Luqi Wang, Fei Gao, Fengyu Cai, Shaojie Shen
ROBIO 2018
Invited Talks
Revela: Dense Retriever Learning via Language Modeling
Beyond Contrastive Pairs: Language Modeling as a Learning Signal for Dense Retrieval
Enhancing Retriever Generalization for Scientific Domain through Complementary Granularity
Academic Service
Media Coverage
ACL 2026 SAC Highlight Oral with Perfect Scores: CoQuIR Teaches Code Retrievers to Pick Good Code
ICLR 2026 Oral | Revela: Redefining Dense Retriever Training via Language Modeling
Employment
Long-horizon agentic RL for LLM post-training on enterprise data.
NLP pipelines for insider-trading risk detection from financial news.
Semi-automatic labeling of insurance contracts with BERT and InferSent.
Education
Supervisors: Heinz Koeppl and Iryna Gurevych. Expected graduation: end of 2026.
Supervisor: Boi Faltings. Final grade: 5.59/6.
Final grade: 3.744/4.3.