Research
Two laboratories working on the foundations of agentic systems and on how people and AI actually get along — with results published openly and tested on real university workloads.
Middle East Technical University
Artificial intelligence, built and studied at METU.
The Department of AI and Data Science at the METU Computer Center brings together research on autonomous agents and human–AI interaction with the infrastructure and services that put artificial intelligence to work across the university.
2research laboratories
On‑campusmodels, data stays at METU
Opento collaboration across faculties
About
A department for AI, inside the university's technical core
AI and Data Science is a department of the METU Computer Center (BİDB). That position is deliberate: we sit where the university's systems, data and networks already run, which lets research move into production and lets production problems shape research.
Infrastructure
GPU compute, virtualization and model-serving operated on campus, so that research groups and administrative units can use modern AI without sending institutional data off site.
Service to the university
Practical AI capability for METU: shared services, technical guidance for faculties and units, and support for teaching and student projects that need real compute.
Laboratories
Where the research happens
Each lab is small, focused, and open to students and collaborators from across the university.
Agent OS
Lab lead: Ramazan Polat · rpolat@metu.edu.tr
Autonomous AI agents increasingly need what programs have always needed from an operating system: a place to run, memory that persists, identity and permissions, and a way to coordinate with others without stepping on each other.
The Agent OS lab studies and builds these foundations — agent runtimes, persistent memory, capability and permission models, auditability, and the protocols by which many agents share one environment safely.
Human–AI Interaction
Lab lead: Ramazan Polat · rpolat@metu.edu.tr
As AI systems move from tools to counterparts, the interesting questions stop being purely technical. How do people build or withdraw trust in an AI system? What do they delegate, and what do they refuse to?
This lab studies the social side of artificial intelligence: interaction and communication norms between people and AI systems, the effects of agentic systems on teams and institutions, and what these systems should be designed to do — and not do — in a university setting.
People
Who works here
The department's research staff. Each entry links to an institutional e-mail address and, where one exists, to the person's page in METU's academic record system (AVESİS).
Özgür Kaya
Principal Investigator
Instructor (Öğr. Gör.) · Department of Computer Engineering, Faculty of Engineering
Ramazan Polat
Lab Lead
Lead, Agent OS and Human–AI Interaction laboratories
Nurbanu Aksoy
Instructor, PhD
Kadir Cenk Alpay
Instructor, PhD
Projects
What the department builds
Systems developed and operated by the department, running on METU infrastructure and used by the METU community.
PRISM
METU's AI learning management system. PRISM turns course material into interactive learning experiences: instructors build subject libraries, connect courses with ODTÜClass, create learning tracks and evaluate student work; students sign in with their METU credentials to browse course libraries, practise with flashcards and assessments, and follow video lessons.
The platform is in its early stages and under active development, and feedback from instructors and students shapes what is built next.
GENTOS
The working system of the Agent OS laboratory. GENTOS gives autonomous agents what programs have always needed from an operating system: a place to run, memory that persists between sessions, identity and permission models, auditability, and protocols that let many agents share one environment without stepping on each other.
In development — the architecture is being built out, and neither the interfaces nor the guarantees are settled yet.
MemHouse
Agent memory as a SQL primitive. MemHouse keeps an agent's memory in a column-oriented database and exposes it through ordinary SQL, so writing, recalling and auditing what an agent remembers needs no bespoke API — any client that speaks SQL can query it.
Because access is a database concern, isolation between agents is enforced by roles and row policies rather than by application code, and every read and write is inspectable after the fact.
In beta — the design is settled and running, and the interfaces are still moving.
Turkish Native LLM
A large language model built natively for Turkish rather than adapted from an English-centric one. The work spans Turkish training data, tokenisation suited to an agglutinative language, and evaluation against tasks that matter in Turkish.
Trained and served on the university's own GPU infrastructure, so both the model and the data behind it stay at METU.
Research and practice
Areas we work in
Our work runs from the theoretical to the immediately operational, and the two ends inform each other.
Agentic systemsRuntimes, memory, tool use and multi-agent coordination.
Language model infrastructureServing, evaluating and operating open-weight models on university hardware.
Human–AI interactionTrust, delegation, and the social effects of AI systems.
Data science and engineeringAnalytical pipelines and data platforms for institutional data.
AI safety and governanceAccess control, auditability and responsible-use policy for AI at scale.
Applied AI for the universityBringing AI to teaching, research support and administrative services.
Services
AI capability, run on campus
We operate shared infrastructure so that researchers, units and students can use modern AI without procuring their own hardware — and without institutional data leaving METU.
METU language model service
Open-weight large language models served on university hardware, through an OpenAI-compatible API and a web chat interface. Prompts and documents are processed on campus and are not sent to any external provider.
Currently in pilot with a limited group of users, and reachable from the campus network.
GPU compute for research
Access to GPU servers for inference and fine-tuning work that does not fit on a desktop machine. Suitable for research groups, graduate projects and course work with a genuine compute requirement.
Requests are reviewed case by case — please get in touch with a short description of your work.
Hosting for research services
Virtual machines and containers for research applications, datasets and internal tools, operated on the department's virtualization platform with institutional networking and TLS.
Technical guidance
Help for faculties and administrative units evaluating AI for a specific problem: what is feasible today, what it costs, where the data-protection line runs, and whether an AI system is the right answer at all.
| Serving | vLLM, open-weight models, OpenAI-compatible endpoints |
|---|---|
| Interfaces | /v1/chat/completions, /v1/completions, /v1/models, web chat |
| Data handling | All processing on METU-operated hardware; no third-party model providers |
Access. Services are available to METU staff, researchers and students. The language model service is currently reachable from the campus network; off-campus access is planned. Write to us to be added to the pilot.
Collaboration
Working with us
We are open to collaboration inside and outside the university.
For METU researchers
Joint projects, compute for AI-heavy research, and help turning a working prototype into a service the university can actually run.
For students
Graduate and undergraduate project topics in both laboratories, plus access to the infrastructure needed to attempt them seriously.
For industry and institutions
Applied research partnerships, technical consultation, and joint work with companies and public institutions on real deployment problems.
Contact
Get in touch
For service access, collaboration proposals, student projects or press enquiries.
- Department
- AI and Data Science
- Unit
- Computer Center (BİDB)
- Institution
- Middle East Technical University
Üniversiteler Mah. Dumlupınar Blv. No:1
06800 Çankaya, Ankara, Türkiye
- ai@metu.edu.tr
- Service requests
- Please include your unit, a short description of the work, and the resources you expect to need.
- METU Computer Center
- bidb.metu.edu.tr