Ahmed Elgohary Ghoneim

Email:

I am a Senior Researcher at Microsoft Research AI Frontiers focusing on training and evaluating agentic models. Recently, I drove the training of MagenticBrain (blog post), an orchestration model trained for multi-step tool calling, coding and terminal use, and delegation to specialized subagents. I also built the team's pipeline for synthesizing terminal environments and tasks, which powered the model's RL stage. Before then, I was part of Microsoft Responsible AI team where I focused on the safety alignment and evaluation of LLMs and agents systems.

I completed my Ph.D. in computer science at the University of Maryland (UMD) in 2021, advised by Jordan Boyd-Graber. My dissertation focused on human-in-the-loop question answering and semantic parsing. Along the way, I was an IBM Ph.D. Fellow and interned at Microsoft Research, AI2, and IBM Research. I also have a soft spot for systems & infra work: at IBM, we got large-scale ML to run directly on compressed data and walked away with the VLDB 2016 Best Paper Award.

Publications

Google Scholar Semantic Scholar

* indicates equal contributions

Blind Spots: Where Small Language Models Fall Behind Frontier Models on Terminal Tasks.
Ashima Suvarna, Ruta Desai* and Ahmed Elgohary.*
The Workshop on Resource-Aware Agentic AI at NeurIPS 2026.
When a Fixed Task Generator Falls Behind: Towards Adaptive Curricula for Training Terminal Agents.
Dang Nguyen, Ruta Desai* and Ahmed Elgohary.*
The Workshop on Towards Test-Time Continual Learning Agents at NeurIPS 2026.
SIRAJ: Diverse and Efficient Red-Teaming for LLM Agents via Distilled Structured Reasoning.
Kaiwen Zhou, Ahmed Elgohary, A S M Iftekhar and Amin Saied.
EACL - Findings 2026.
An earlier version appears at The Lock-LLM Workshop at NeurIPS 2025.
Jailbreak Distillation: Renewable Safety Benchmarking.
Jingyu Zhang, Ahmed Elgohary, Xiawei Wang, A S M Iftekhar, Ahmed Magooda, Benjamin Van Durme, Daniel Khashabi and Kyle Jackson.
EMNLP - Findings 2025.
Also appears at The Lock-LLM Workshop at NeurIPS 2025.
Controllable Safety Alignment: Inference-Time Adaptation to Diverse Safety Requirements.
Jingyu Zhang, Ahmed Elgohary, Ahmed Magooda, Daniel Khashabi and Benjamin Van Durme.
ICLR 2025.
An earlier version appears at The Pluralistic Alignment Workshop at NeurIPS 2024.
Human-in-the-Loop Question Answering with Natural Language Interaction
Ahmed Elgohary
Ph.D. Dissertation, University of Maryland, 2021.
NL-EDIT: Correcting semantic parse errors through natural language interaction.
Ahmed Elgohary, Christopher Meek, Matthew Richardson, Adam Fourney, Gonzalo Ramos and Ahmed Hassan Awadallah
NAACL 2021.
Speak to your parser: Interactive Text-to-SQL with natural language feedback.
Ahmed Elgohary, Saghar Hosseini and Ahmed Hassan Awadallah.
ACL 2020.
It takes two to lie: One to lie, and one to listen.
Denis Peskov, Benny Cheng, Ahmed Elgohary, Joe Barrow, Cristian Danescu-Niculescu-Mizil and Jordan Boyd-Graber.
ACL 2020.
Can you unpack that? Learning to rewrite questions-in-context.
Ahmed Elgohary, Denis Peskov, Jordan Boyd-Graber.
EMNLP 2019.
Dataset and baselines for sequential open-domain question answering
Ahmed Elgohary,* Chen Zhao,* Jordan Boyd-Graber.
EMNLP 2018.
Generating natural language adversarial examples.
Moustafa Alzantot, Yash Sharma, Ahmed Elgohary, Bo-Jhang Ho, Mani Srivastava, Kai-Wei Chang.
EMNLP 2018.
Assessing composition in sentence vector representations.
Allyson Ettinger, Ahmed Elgohary, Colin Phillips, Philip Resnik.
COLING 2018.
Construction of the literature graph in semantic scholar.
Waleed Ammar, Dirk Groeneveld, Chandra Bhagavatula, Iz Beltagy, Miles Crawford, Doug Downey,
Jason Dunkelberger, Ahmed Elgohary, Sergey Feldman, Vu Ha, Rodney Kinney, Sebastian Kohlmeier,
Kyle Lo, Tyler Murray, Hsu-Han Ooi, Matthew Peters, Joanna Power, Sam Skjonsberg, Lucy Lu Wang,
Chris Wilhelm, Zheng Yuan, Madeleine van Zuylen, Oren Etzioni.
NAACL 2018.
Probing for semantic evidence of composition by means of simple classification tasks.
Allyson Ettinger, Ahmed Elgohary, Philip Resnik.
RepEval Workshop at ACL 2016. [Best Paper Award]
Learning monolingual compositional representations via bilingual supervision.
Ahmed Elgohary and Marine Carpuat.
ACL 2016.
Compressed linear algebra for large-scale machine learning.
Ahmed Elgohary, Matthias Boehm, Peter J. Haas, Frederick R. Reiss, Berthold Reinwald.
VLDB 2016. [Best Paper Award] [SIGMOD Research Highlights Award]
[IBM Pat Goldberg Memorial Best Paper Award]
Extended in VLDB Journal, 2018 and Communications of the ACM, 2019.
The UMD Machine Translation Systems at IWSLT 2015
Amittai Axelrod, Ahmed Elgohary, Marianna Martindale, Khánh Nguyễn,
Xing Niu, Yogarshi Vyas, Marine Carpuat.
IWSLT 2015.
Scalable embeddings for kernel clustering on MapReduce.
Ahmed Elgohary
Master's Thesis, University of Waterloo, 2014.
Embed and Conquer: Scalable embeddings for kernel k-Means on MapReduce.
Ahmed Elgohary, Ahmed K. Farahat, Mohamed S. Kamel, Fakhri Karray.
SIAM Conference on Data Mining (SDM) 2014.
Distributed column subset selection on MapReduce.
Ahmed K. Farahat, Ahmed Elgohary, Ali Ghodsi, Mohamed S. Kamel.
IEEE International Conference on Data Mining (ICDM) 2013. [Best Paper Award Runner-Up]
Extended in Knowledge and Information Systems, 2014.
Approximate nearest centroid embedding for kernel k-means.
Ahmed Elgohary, Ahmed K. Farahat, Mohamed S. Kamel, Fakhri Karray.
Big Learning Workshop at NeurIPS 2013.
No need to war-drive: Unsupervised indoor localization
He Wang, Souvik Sen, Ahmed Elgohary, Moustafa Farid, Moustafa Youssef, Romit Roy Choudhury.
ACM International Conference on Mobile Systems (MobiSys) 2012.
Extended in IEEE Transactions on Mobile Computing, 2016.