akougkas.io

Anthony Kougkas, PhD

I build IOWarp and CLIO, context infrastructure for scientific AI at the Gnosis Research Center, and the line of HPC storage and I/O systems that led to them. The work is the stretch between a promising idea and something other people can run, question, and carry forward.

Hi. I’m Anthony.

I try to make scientific computers stop waiting around for their data. That is the short version. Today it takes the form of IOWarp and CLIO, context infrastructure for scientific AI that I build at the Gnosis Research Center. The longer version is about turning research questions into systems, turning prototypes into software, and creating enough structure around a good idea that other people can own it.

I was an officer in the Hellenic Army for nearly nine years before I was a computer scientist, in environments where reliability was not an abstract quality. When systems failed, people noticed. The lesson stayed: reliability is the floor, and features are built on top of it.

Chicago became home when I came to Illinois Tech for a PhD under Xian-He Sun. The question I settled into sounded narrow: how do we get data to processors without wasting the machine around them? It kept opening into larger questions about storage, metadata, context, tools, and the people using them.

I am most interested in the moment when an idea has to negotiate with the world: dependencies, users, collaborators, evidence, failure, and change.

The road to IOWarp and CLIO

My research is one line of questions, each opened by the answer before it. IRIS asked whether files and objects could sit behind one interface. Hermes asked who should decide where data lives once a machine has many tiers of memory and storage, and made placement the middleware’s job. LABIOS asked what that middleware could do if each request said what it was for, and made the label carry intent. ChronoLog asked how to order the activity those requests leave behind, and chose physical time over consensus. DeepIO, my series of subcontracts with Argonne, Pacific Northwest, and Lawrence Livermore, asked what deep learning needs from the I/O stack, and funded the students who followed that question from training to inference. DTIO treated the label as a unit of work, a data task a runtime can compose and schedule. IOWarp asks the question all of them were circling: which data, metadata, tools, and history does a computation need right now? IOWarp carries Hermes placement, LABIOS intent, DTIO tasks, and ChronoLog provenance forward, and CLIO is the architecture that gives that context to scientific AI agents.

LABIOS won the Karsten Schwan Best Paper Award at HPDC 2019 and became US Patent 11,630,834. DLIO, the benchmark Hariharan Devarajan built with Argonne while I was part of the team co-advising him, won the CCGrid 2021 Best Paper Award.

NSF funded IOWarp in 2024 as a collaboration with The HDF Group and the University of Utah. The research direction, component boundaries, shared contracts, integration, partnerships, and releases across the ecosystem are mine to answer for. Luke Logan leads day-to-day engineering in CLIO Core. Jaime Cernuda leads CLIO Agent and Relay development. Students build CLIO Kit, and research engineers and students develop the other components. I contribute across the stack and create and maintain CLIO Coder myself, as one hands-on part of the program.

What I build

Several threads keep reconnecting:

  • Storage hierarchies should be steerable by applications rather than invisible machinery.
  • I/O should carry intent instead of becoming an anonymous byte stream.
  • Benchmarks are tools for arguing honestly about behavior.
  • AI systems need the right data, metadata, tools, computational state, and human decisions at the right moment.
  • Research becomes more useful when someone can run, inspect, and extend it.

How I work

Build beyond the paper.

A result becomes more useful when somebody can install it, test it, disagree with it, or carry it into a different system.

Make the evidence legible.

Code, releases, benchmarks, papers, talks, and limitations should tell one connected story.

Give people ownership.

Mentoring works when a person owns the decisions as well as the tasks that follow from them.

Say what is not true yet.

Say what works, what is experimental, what is planned, and what remains uncertain.

The people I work with

Xian-He and I co-founded the Gnosis Research Center at Illinois Tech in 2023, when the work had outgrown one faculty lab. Ambitious systems work takes longer than a semester and involves more than a paper, and GRC keeps the repositories, machines, proposals, and collaborations in place long enough for it to compound. Building GRC tells how it grew.

The implementation is collective. Students, research staff, and collaborators lead parts of the systems, challenge the architecture, write the papers, build the software, and carry the work into new places. Luke and Jaime both did their PhDs inside these systems and stayed on at GRC as research faculty in January 2026.

Alumni:

My aim as a mentor is independent judgment. I want people to grow past my advice, disagree with an architecture for reasons I missed, and have the evidence and communication skill to carry a better idea forward.

Collaborators and institutions

The systems above were built in long collaborations. Xian-He Sun has worked with me across Hermes, LABIOS, ChronoLog, and IOWarp since my PhD. Partners at national laboratories, universities, and The HDF Group carried those projects with me as coauthors and co-investigators.

Current roles

Connect

Write if you are building around data systems, scientific AI, research software, or agent infrastructure; if a workload is slower or stranger than it should be; or if something here gives you a useful disagreement. If you are a student thinking about a PhD in data systems, email me about a PhD and say what you have built and what you want to break. The teaching record and the students I co-advise show what that looks like.

The direct public channel is akougkas@illinoistech.edu. The formal record is in the Archive and the CV.