Ansel K. Erol

San Francisco, CA

I’m a FDE at Baseten with a focus on deploying real-time, high-throughput software systems at scale for interactive model serving and artificial intelligence workloads.

My research at the Georgia Tech Systems for Scalable Intelligence (SCAI) Lab (advised by Dr. Divya Mahajan) centered on scaling ML inference on constrained and distributed systems, including efficient MoE model serving and edge computing on satellites. I have presented first-author work at MLSys ‘25 and ‘26, and received the Georgia Tech Outstanding MS Research Award and the Marshall D. Williamson Fellowship for academic excellence and effectiveness as a graduate teaching assistant for CS3600: Intro to AI.

I previously worked at Google (Mountain View, P&D Privacy Sandbox), where I optimized gRPC networking for a low-latency C++ inference system in private ML ad infrastructure. I am also a Board Member and former Finance Director of People for PSEO, a Minnesota non-profit advocating for dual-enrollment students.

Outside of work and research, I enjoy classical music, travel, pastry, and poetry — visit my passions page to learn more.

News

Sep 04, 2026	Our paper Why Do AI Agents Break Rules? How Framing, Context, and Social Signals Shape Compliance was accepted to the AAAI/ACM Conference on AI, Ethics, and Society (AIES) 2026! Explore the interactive results.
Apr 05, 2026	Thrilled to announce I’ll be joining Baseten as a Forward Deployed Software Engineer this June!
Apr 01, 2026	Honored to receive the Marshall D. Williamson Fellowship and the Georgia Tech Outstanding MS Research Award from the Georgia Tech College of Computing!
Mar 08, 2026	Our paper Explainable Model Routing for Agentic Workflows was accepted as a Spotlight at the ACM CHI 2026 Human-Centered Explainable AI (HCXAI) Workshop!
Feb 12, 2026	Presented EarthSight at the CRNCH Summit 2026, and it has been accepted to appear at the 2026 Annual Conference on Machine Learning and Systems (MLSys 2026)! Slides: https://mlsys.org/media/mlsys-2026/Slides/3792.pdf

Selected Publications

EarthSight: A Distributed Framework for Low-Latency Satellite Intelligence

A. K. Erol, S. Lee, and D. Mahajan

In Annual Conference on Machine Learning and Systems (MLSys), 2026

To appear. Also presented at GT Undergraduate Research Symposium (Apr 2025), MLSys Young Professionals Symposium (May 2025), and CRNCH Summit (Feb 2026)

First Author arXiv Bib Slides

Presented as first author at MLSys ’25 and ’26

@inproceedings{erol2025earthsight,
  title = {EarthSight: A Distributed Framework for Low-Latency Satellite Intelligence},
  author = {Erol, A. K. and Lee, S. and Mahajan, D.},
  booktitle = {Annual Conference on Machine Learning and Systems (MLSys)},
  year = {2026},
  note = {To appear. Also presented at GT Undergraduate Research Symposium (Apr 2025), MLSys Young Professionals Symposium (May 2025), and CRNCH Summit (Feb 2026)},
}

PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?

M. Okamoto and A. K. Erol

2026

Under review

arXiv Bib Code Data Website

@misc{okamoto2026pact,
  title = {{PACT}: Can Enterprise {AI} Assistants Be Trusted Under Pressure?},
  author = {Okamoto, M. and Erol, A. K.},
  year = {2026},
  howpublished = {Preprint},
  note = {Under review},
  dataset = {https://huggingface.co/datasets/trace-ai-labs/pact}
}

Why Do AI Agents Break Rules? How Framing, Context, and Social Signals Shape Compliance

M. Okamoto, A. K. Erol, and K. Erol

In AAAI/ACM Conference on AI, Ethics, and Society (AIES), 2026

Also presented at the COLM 2026 Workshop on Agent Behavior

arXiv Bib Code Website

@inproceedings{okamoto2026whydo,
  title = {Why Do {AI} Agents Break Rules? How Framing, Context, and Social Signals Shape Compliance},
  author = {Okamoto, M. and Erol, A. K. and Erol, K.},
  booktitle = {AAAI/ACM Conference on AI, Ethics, and Society (AIES)},
  year = {2026},
  note = {Also presented at the COLM 2026 Workshop on Agent Behavior},
}

Explainable Model Routing for Agentic Workflows

M. Okamoto, A. K. Erol, and M. Riedl

2026

Spotlight

Spotlight arXiv Bib

Spotlight at ACM CHI HCXAI Workshop 2026

@misc{okamoto2026explainablemodelroutingagentic,
  title = {Explainable Model Routing for Agentic Workflows},
  author = {Okamoto, M. and Erol, A. K. and Riedl, M.},
  year = {2026},
  howpublished = {ACM CHI 2026 Human-Centered Explainable AI (HCXAI) Workshop},
  note = {Spotlight},
}

Trust by Design: Skill Profiles for Transparent, Cost-Aware LLM Routing

M. Okamoto, A. K. Erol, and G. Matlin

In 8th Annual Conference on Machine Learning and Systems, Young Professionals Symposium, 2025

arXiv Bib

@inproceedings{okamoto2025trust,
  title = {Trust by Design: Skill Profiles for Transparent, Cost-Aware {LLM} Routing},
  author = {Okamoto, M. and Erol, A. K. and Matlin, G.},
  booktitle = {8th Annual Conference on Machine Learning and Systems, Young Professionals Symposium},
  year = {2025},
}

Equinox: Decentralized Scheduling for Hardware-Aware Orbital Intelligence

A. K. Erol

In Annual Conference on Machine Learning and Systems (MLSys), Young Professionals Symposium, 2026

Presented at MLSys 2026 Young Professionals Symposium; MS thesis

arXiv Bib PDF Poster

@inproceedings{erol2026equinox,
  title = {Equinox: Decentralized Scheduling for Hardware-Aware Orbital Intelligence},
  author = {Erol, A. K.},
  booktitle = {Annual Conference on Machine Learning and Systems (MLSys), Young Professionals Symposium},
  year = {2026},
  note = {Presented at MLSys 2026 Young Professionals Symposium; MS thesis},
}

A Holistic Approach to Workforce Scheduling and Routing

C. K. Manalp, A. K. Erol, K. Erol, and 1 more author

Turkish Journal of Electrical Engineering and Computer Science, 2025

Accepted, to appear

Bib

@article{manalp2025workforce,
  title = {A Holistic Approach to Workforce Scheduling and Routing},
  author = {Manalp, C. K. and Erol, A. K. and Erol, K. and Evrendilek, C.},
  journal = {Turkish Journal of Electrical Engineering and Computer Science},
  year = {2025},
  note = {Accepted, to appear},
}