Milos Gravara

Research Scientist

BSc, MSc

m.gravara@dsg.tuwien.ac.at

+43 667 3312173

Short CV

Milos Gravara is a PhD student and Project Assistant at the Distributed Systems Group at TU Wien. His research focuses on Distributed and Compound AI Systems, with particular emphasis on AI inference optimization, performance modeling and optimization, and the execution of AI workloads across heterogeneous computing environments.

He received his MSc in Electrical and Computer Engineering from the Faculty of Technical Sciences, University of Novi Sad in 2024, specializing in High-Performance Computing. In January 2025, he joined the Distributed Systems Group of the Institute of Information Systems at TU Wien to pursue his PhD.

Research Interests

Distributed and Compound AI Systems. Design, deployment, and operation of distributed AI applications and Compound AI workflows that combine multiple AI models, tools, and software components, including Agentic AI and Mixture-of-Experts architectures, across heterogeneous computing infrastructure.
AI Inference Optimization. Techniques for efficient AI inference and serving, including model selection, resource allocation, heterogeneous placement, and optimization of latency, throughput, accuracy, and cost.
AI Performance Modeling and Optimization. Methods for modeling and estimating the end-to-end performance and accuracy of AI workflows, exploring their configuration spaces, and optimizing configurations and deployments under SLOs and resource constraints.
Serverless Computing. Serverless execution and resource management for distributed and AI workloads, including workflow orchestration, memory management, elastic serving, and scale-to-zero execution.
Edge-Cloud-Space Continuum and Space Data Centers. Scheduling, placement, and orchestration of AI workloads across terrestrial and orbital computing infrastructure, with a focus on heterogeneous resources, dynamic connectivity and topology, and emerging space data centers.

Thesis Supervision

If you are interested in one of the topics above, I can co-advise your bachelor’s or master’s thesis.

Master’s Theses

Manuel Janisch. Comparative Analysis of Reinforcement Learning Approaches for Runtime Model Selection in Compound AI Systems. (in progress)
Nico Kratky. Serverless Scale-to-Zero Serving for Mixture-of-Experts Models. (in progress)

Bachelor’s Theses

Darya Haponava. Serverless Orchestration and Memory Management for Multi-Agent LLM Workflows. (completed)

Projects

NexaSphere

NexaSphere is a Horizon Europe research and innovation project focused on designing and validating a unified three-dimensional communication network integrating terrestrial, non-terrestrial, and aerial systems for future 6G connectivity. Within this broader scope, the project includes research on distributed computing, AI-enabled resource management, and orchestration across the edge-cloud-space continuum.

Publications

M. Gravara, A. Stanisic, and S. Nastic, “Atlas: Optimizing Deployment of Compound AI Workflows on Heterogeneous Clusters,” in Proc. IEEE/ACM Symposium on Edge Computing (SEC), 2026.
A. Stanisic, M. Gravara, J. L. Herrera, and S. Nastic, “Constella: A Novel Framework for Cost-Efficient Distributed AI Inference in LEO Space Data Centers,” in Euro-Par 2026: Parallel Processing, 32nd European Conference on Parallel and Distributed Processing, Proceedings, Part II, pp. 347–362, 2026.
M. Gravara, A. Stanisic, and S. Nastic, “Design Methodology and Performance Trade-offs Management for Distributed and Compound AI Systems,” in Proc. IEEE International Conference on Cloud Computing (CLOUD), IEEE World Congress on SERVICES, 2026.
M. Gravara, C. Marcelino, A. Stanisic, and S. Nastic, “PLAIground: SLO-Driven Runtime Model Selection for Compound AI Systems in the Edge-Cloud-Space Continuum,” in Proc. IEEE International Conference on Smart Computing Workshops and Other Affiliated Events (SmartComp Companion), pp. 261–266, 2026.
M. Gravara, J. L. Herrera, and S. Nastic, “Compass: Optimizing Compound AI Workflows for Dynamic Adaptation,” in Proc. IEEE 26th International Symposium on Cluster, Cloud and Internet Computing (CCGrid), pp. 84–93, 2026.
M. Gravara, A. Stanisic, and S. Nastic, “A Novel Compound AI Model for 6G Networks in 3D Continuum,” in Proc. European Conference on Networks and Communications & 6G Summit (EuCNC & 6G Summit), 2025.