Mehmet Onur Keskin

Amsterdam, the Netherlands

I study how autonomous agents make complex decisions when goals conflict, information is incomplete, and environments dynamically shift. My research spans negotiation systems that model and respond to human preferences and emotions, decentralized cooperation and orchestration among agents, and adaptive information retrieval. I develop and evaluate systems for real-time use, drawing on experience building research frameworks and leading agile engineering teams that brought AI systems into production at scale.

I completed my M.Sc. in Artificial Intelligence, graduating Summa Cum Laude, under the supervision of Dr. Reyhan Aydogan at the Interactive Intelligent Systems Lab, Ozyegin University. In my master’s thesis, I investigated multimodal negotiation protocols that enable autonomous agents to model human counterparts’ preferences and negotiation strategies, generate persuasive arguments and gestures, and adapt to verbal and nonverbal affective signals.

In my human-robot negotiation research and our broader work on decision-making under uncertainty, we have repeatedly demonstrated how autonomous agents can solve multi-objective, partially observable problems and reach Pareto-optimal outcomes alongside human partners in real time. This broader work includes multi-agent path planning, supply chain optimization, and information retrieval. These works have resulted in 15 peer-reviewed publications, including conference papers at IJCAI, EMNLP and AAMAS, journal articles in IEEE THMS and ACM TiiS, and a book chapter. Our project won first place in the IBM Watson Human Multi-Agent Negotiation Challenge at IJCAI 2020. I was selected as a Young Researcher at the 12th Heidelberg Laureate Forum and an IEEE CS Richard Merwin Scholar.

I have been lucky enough to shape my research perspective through close collaborations and research visits:

Interactive Intelligence, TU Delft Host: Prof. Catholijn Jonker

Built multimodal negotiation and behavioral evaluation frameworks (IJCAI 2024). Extending the affective negotiation research from Cambridge, adapted personalized policies across multiple humanoids (Pepper, NAO, and QTrobot).

AFAR Lab, University of Cambridge Host: Prof. Hatice Gunes

Developed a differential affective feedback mechanism on humanoid robots with continual learning, personalizing valence-arousal perception in real time to overcome catastrophic forgetting and adapt negotiation strategies.

E-GLAMOR Lab, Brown University Host: Prof. Amy Greenwald

Designed a supply chain manager agent deploying a 3D combinatorial dispatcher for multi-issue, multi-dealer negotiations, utilizing forecasts and production planning to maximize long-horizon profit-risk ratios (2^nd at IJCAI SCML'21).

CoRIS Institute, Oregon State University Host: Prof. Yigit Menguc

Formulated viscous silicone 3D printing toolpaths as an Asymmetric Traveling Salesperson Problem, evaluating heuristic approaches like Chained Lin-Kernighan for collision-free paths to cut printing time by 34% (NSF REU Scholar).

Beyond academic research, I co-founded Kanzy Financial Technologies in London. Leading an engineering team, I guided the development of Finguru, an agent-native investment platform supported by NVIDIA Inception and the Google for Startups Cloud Program. We built real-time streaming pipelines over complex financial disclosures, verifiable multi-agent reasoning workflows, and adaptive retrieval routing across dense and late-interaction vision-language models. Building production systems provided valuable practical insight into how theoretical decision models behave when deployed for real users. I have stepped down from my operational role at Kanzy to return to foundational AI research.

Recent News

Aug 2026	Our paper “RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval” was accepted to the Conference on Empirical Methods in Natural Language Processing (EMNLP 2026).
Aug 2026	I have since stepped down from my operational role at Kanzy Financial Technologies to focus on foundational AI research.

Selected Publications

IEEE TAC

Personalized Affective Perception via Continual Learning in Human-Agent Negotiation [Under Review]

Mehmet Onur Keskin, Berk Buzcu, Nikhil Churamani, and

IEEE Transactions on Affective Computing | IF: 11.3

Submission: May 2026

EMNLP 2026

RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval

Emre Kuru, Mehmet Onur Keskin, Reza Farahbakhsh, and

In Conference on Empirical Methods in Natural Language Processing (EMNLP), Main Conference

Oct 2026 | Acceptance Rate: %15.39 (2,719/17,669)

PDF arXiv Code

Document retrieval increasingly supports high-stakes information access in finance, healthcare, and law. Modern retrieval pipelines vary both in modality (text or multimodal) and in retrieval architecture (dense or late-interaction). These choices impose a hard compromise: the most effective pipelines are too slow and expensive to run at scale, while the fastest fail to retrieve evidence from complex documents. Practitioners must therefore choose between missed evidence and unusable latency, with no principled basis for adapting that choice at the query level. We show that this compromise is unnecessary. Not every query requires the same pipeline. Across benchmarks spanning financial and scientific corpora, no static pipeline dominates. We introduce RetrievalRouter, a lightweight query-aware router that learns, from the query text alone, which retrieval pipeline best fits each query. A single tunable parameter exposes the full accuracy-latency frontier, and for every static baseline, RetrievalRouter offers an operating point that is simultaneously more accurate and faster. Against the best static baseline, RetrievalRouter is 2.5% more accurate and 12.4 times faster. Furthermore, compared with prior adaptive strategy selection methods, RetrievalRouter achieves significantly higher nDCG@5 across accuracy-oriented settings, while matching or numerically outperforming them on both nDCG@5 and latency in latency-oriented settings. Our code and data are available at https://github.com/emrekuruu/retrieval-router.

@inproceedings{kuru2026retrievalrouter,
  month = oct,
  note = {Acceptance Rate: %15.39 <b>(2,719/17,669)</b>; Accepted},
  url = {https://arxiv.org/abs/2608.25625},
  author = {Kuru, Emre and Keskin, Mehmet Onur and Farahbakhsh, Reza and Crespi, Noel},
  title = {RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval},
  booktitle = {Conference on Empirical Methods in Natural Language Processing (EMNLP), Main Conference},
  year = {2026},
  abstract = {Document retrieval increasingly supports high-stakes information access in finance, healthcare, and law. Modern retrieval pipelines vary both in modality (text or multimodal) and in retrieval architecture (dense or late-interaction). These choices impose a hard compromise: the most effective pipelines are too slow and expensive to run at scale, while the fastest fail to retrieve evidence from complex documents. Practitioners must therefore choose between missed evidence and unusable latency, with no principled basis for adapting that choice at the query level. We show that this compromise is unnecessary. Not every query requires the same pipeline. Across benchmarks spanning financial and scientific corpora, no static pipeline dominates. We introduce RetrievalRouter, a lightweight query-aware router that learns, from the query text alone, which retrieval pipeline best fits each query. A single tunable parameter exposes the full accuracy-latency frontier, and for every static baseline, RetrievalRouter offers an operating point that is simultaneously more accurate and faster. Against the best static baseline, RetrievalRouter is 2.5% more accurate and 12.4 times faster. Furthermore, compared with prior adaptive strategy selection methods, RetrievalRouter achieves significantly higher nDCG@5 across accuracy-oriented settings, while matching or numerically outperforming them on both nDCG@5 and latency in latency-oriented settings. Our code and data are available at https://github.com/emrekuruu/retrieval-router.}
}

IJCAI 2024 Demo Track

NEGOTIATOR: A Comprehensive Framework for Human-Agent Negotiation Integrating Preferences, Interaction, and Emotion

Mehmet Onur Keskin, Berk Buzcu, Berkecan Koçyiğit, and

In Proceedings of the 33rd International Joint Conference on Artificial Intelligence (IJCAI), Jeju, Korea

Aug 2024 | Acceptance Rate: %18.28

DOI PDF Code

The paper introduces a comprehensive human-agent negotiation framework designed to facilitate the development and evaluation of research studies on human-agent negotiation without building each component from scratch. Leveraging the interoperability and reusability of its components, this framework offers various functionalities, including speech-to-text conversion, emotion recognition, a repository of negotiation strategies, and an interaction manager capable of managing gestures designed for Nao, Pepper, and QT, and coordinating message exchanges in a turn-taking fashion. This framework aims to lower the entry barrier for researchers in human-agent negotiation by providing a versatile platform that supports a wide range of research directions, including affective computing, natural language processing, decision-making, and non-verbal communication.

@inproceedings{10.24963/ijcai.2024/1012,
  pages = {8700--8703},
  month = aug,
  note = {Acceptance Rate: %18.28},
  author = {Keskin, Mehmet Onur and Buzcu, Berk and Koçyiğit, Berkecan and Çakan, Umut and Doğru, Anıl and Aydoğan, Reyhan},
  booktitle = {Proceedings of the 33rd International Joint Conference on Artificial Intelligence (IJCAI), Jeju, Korea},
  title = {NEGOTIATOR: A Comprehensive Framework for Human-Agent Negotiation Integrating Preferences, Interaction, and Emotion},
  abstract = {The paper introduces a comprehensive human-agent negotiation framework designed to facilitate the development and evaluation of research studies on human-agent negotiation without building each component from scratch. Leveraging the interoperability and reusability of its components, this framework offers various functionalities, including speech-to-text conversion, emotion recognition, a repository of negotiation strategies, and an interaction manager capable of managing gestures designed for Nao, Pepper, and QT, and coordinating message exchanges in a turn-taking fashion. This framework aims to lower the entry barrier for researchers in human-agent negotiation by providing a versatile platform that supports a wide range of research directions, including affective computing, natural language processing, decision-making, and non-verbal communication.},
  year = {2024},
  isbn = {978-1-956792-04-1},
  url = {https://doi.org/10.24963/ijcai.2024/1012},
  doi = {10.24963/ijcai.2024/1012},
  articleno = {1012},
  numpages = {4},
  series = {IJCAI '24}
}

JAAMAS 2024

Recorded DEC-MAPF simulation with 80 agents on a 32 by 32 obstacle map, following agent 2 with a white ring and cyan field of view.

Decentralized Multi-Agent Path Finding Framework and Strategies Based on Automated Negotiation

Mehmet Onur Keskin, Furkan Cantürk, Cihan Eran, and

Autonomous Agents and Multi-Agent Systems | IF: 2.4

Jun 2024

DOI PDF Code Docs

This paper introduces a negotiation framework to solve the Multi-Agent Path Finding (MAPF) Problem for self-interested agents in a decentralized fashion. The framework aims to achieve a good trade-off between the privacy of the agents and the effectiveness of solutions. Accordingly, a token-based bilateral negotiation protocol and two negotiation strategies are presented. The experimental results over four different settings of the MAPF problem show that the proposed approach could find conflict-free path solutions albeit suboptimally, especially when the search space is large and high-density. In contrast, Explicit Estimation Conflict-Based Search (EECBS) struggles to find optimal solutions. Besides, deploying a sophisticated negotiation strategy that utilizes information about local density for generating alternative paths can yield remarkably better solution performance in this negotiation framework.

@article{Keskin2024_MAPF,
  articleno = {10},
  number = {1},
  volume = {38},
  month = jun,
  journal = {Autonomous Agents and Multi-Agent Systems},
  author = {Keskin, Mehmet Onur and Cantürk, Furkan and Eran, Cihan and Aydoğan, Reyhan},
  title = {Decentralized Multi-Agent Path Finding Framework and Strategies Based on Automated Negotiation},
  year = {2024},
  doi = {10.1007/s10458-024-09639-8},
  url = {https://doi.org/10.1007/s10458-024-09639-8},
  abstract = {This paper introduces a negotiation framework to solve the Multi-Agent Path Finding (MAPF) Problem for self-interested agents in a decentralized fashion. The framework aims to achieve a good trade-off between the privacy of the agents and the effectiveness of solutions. Accordingly, a token-based bilateral negotiation protocol and two negotiation strategies are presented. The experimental results over four different settings of the MAPF problem show that the proposed approach could find conflict-free path solutions albeit suboptimally, especially when the search space is large and high-density. In contrast, Explicit Estimation Conflict-Based Search (EECBS) struggles to find optimal solutions. Besides, deploying a sophisticated negotiation strategy that utilizes information about local density for generating alternative paths can yield remarkably better solution performance in this negotiation framework.}
}

IEEE THMS 2022

Would You Imagine Yourself Negotiating With a Robot, Jennifer? Why Not?

Reyhan Aydoğan, Mehmet Onur Keskin, and Umut Çakan

IEEE Transactions on Human-Machine Systems | IF: 4.4

Feb 2022

DOI PDF Code

With the improvement of intelligent systems and robotics, social robots are becoming part of our society. To accomplish complex tasks, robots and humans may need to collaborate, and when necessary, they need to negotiate with each other. While designing such socially interacting robots, it is crucial to consider human factors such as facial expression, emotions, and body language. Since gestures play a crucial role in interaction, this article studies the effect of gestures in human–robot negotiation experiments. Additionally, it compares the performance of variants of the well-known negotiation tactics (i.e., time-based and behavior-based) in automated negotiation literature in the context of human–robot negotiations. Our experimental results support the finding in automated negotiation. That is, the robot gained higher utility when it imitates its opponent’s bidding strategy than employing a time-based negotiation strategy. When adopting a behavior-based technique, there is a statistically significant effect of gestures on the underlying negotiation process, and, therefore, on negotiation outcome.

@article{9609976,
  month = feb,
  journal = {IEEE Transactions on Human-Machine Systems},
  author = {Aydoğan, Reyhan and Keskin, Mehmet Onur and Çakan, Umut},
  title = {Would You Imagine Yourself Negotiating With a Robot, Jennifer? Why Not?},
  abstract = {With the improvement of intelligent systems and robotics, social robots are becoming part of our society. To accomplish complex tasks, robots and humans may need to collaborate, and when necessary, they need to negotiate with each other. While designing such socially interacting robots, it is crucial to consider human factors such as facial expression, emotions, and body language. Since gestures play a crucial role in interaction, this article studies the effect of gestures in human–robot negotiation experiments. Additionally, it compares the performance of variants of the well-known negotiation tactics (i.e., time-based and behavior-based) in automated negotiation literature in the context of human–robot negotiations. Our experimental results support the finding in automated negotiation. That is, the robot gained higher utility when it imitates its opponent’s bidding strategy than employing a time-based negotiation strategy. When adopting a behavior-based technique, there is a statistically significant effect of gestures on the underlying negotiation process, and, therefore, on negotiation outcome.},
  year = {2022},
  volume = {52},
  number = {1},
  pages = {41-51},
  doi = {10.1109/THMS.2021.3121664}
}

AAMAS 2021

Solver Agent: Towards Emotional and Opponent-Aware Agent for Human-Robot Negotiation

Mehmet Onur Keskin, Umut Çakan, and Reyhan Aydoğan

In International Conference on Autonomous Agents and Multiagent Systems

May 2021 | Acceptance Rate: %39.17

DOI PDF Code

Negotiation is one of the crucial processes for resolving conflicts between parties. In automated negotiation, agent designers mostly take opponent’s offers and the remaining time into account while designing their strategies. While designing a negotiating agent interacting with a human directly, other information such as opponent’s emotional changes during the negotiation can establish a better interaction and reach an admissible settlement for joint interests. Accordingly, this paper proposes a bidding strategy for humanoid robots, which incorporates their opponents’ emotional states and awareness of the agent’s changing behavior.

@inproceedings{inproceedings,
  month = may,
  note = {Acceptance Rate: %39.17},
  booktitle = {International Conference on Autonomous Agents and Multiagent Systems},
  url = {https://www.ifaamas.org/Proceedings/aamas2021/pdfs/p1557.pdf},
  author = {Keskin, Mehmet Onur and Çakan, Umut and Aydoğan, Reyhan},
  year = {2021},
  pages = {1557--1559},
  title = {Solver Agent: Towards Emotional and Opponent-Aware Agent for Human-Robot Negotiation},
  abstract = {Negotiation is one of the crucial processes for resolving conflicts between parties. In automated negotiation, agent designers mostly take opponent's offers and the remaining time into account while designing their strategies. While designing a negotiating agent interacting with a human directly, other information such as opponent's emotional changes during the negotiation can establish a better interaction and reach an admissible settlement for joint interests. Accordingly, this paper proposes a bidding strategy for humanoid robots, which incorporates their opponents' emotional states and awareness of the agent's changing behavior.}
}