Dear Readers,

As we reach the conclusion of the MLSysOps project, I look back at this journey and what our consortium has achieved over the past three years of collaborative work.

We began with the vision to tackle one of the most pressing challenges in modern computing: managing the vast, heterogeneous cloud-edge-IoT continuum without constant human supervision and intervention. Today, I am proud to say that we have made very significant progress toward this vision. We have successfully developed and evaluated a complete AI-ready framework, spanning the cloud, edge, IoT and even far-edge layers, which supports the end-to-end deployment of hyper-distributed applications in a largely transparent way, and enables autonomic system management at various layers of the continuum via plug-in policies and flexible engagement of ML models, bringing intelligence directly to where data is born.

In this final edition of our newsletter, we report the most recent developments of the project, focusing on the results achieved through the MLSysOps framework in real-world application scenarios: both our use cases, in smart cities and precision agriculture, demonstrate how autonomic ML-driven application and system management can optimize real-world operations, from energy-efficient urban traffic monitoring via smart embedded nodes equipped with cameras, to improved weed detection in the field via the smart engagement of a drone in tandem with a tractor.

We also report on our Hackathon event, which attracted more than 60 participants, and the second edition of our ML4ECS workshop at HiPEAC 2026. Finally, do not forget to check out our recent publications and the large suite of publicly available artefacts (simulators, software, datasets, ML models), which can be freely used by the community.

Thank you for following our progress.

Spyros Lalis
Coordinator of MLSysOps

We are proud to present the final results of the Smart Agriculture use case within the MLSysOps project, achieved through extensive field validation and outstanding dedication of our team and collaborators. This use case addressed a key challenge in precision agriculture: maintaining accurate, real-time weed detection under rapidly changing field conditions such as sunlight and shadows.

To tackle this, a tractor and an autonomous drone, both running the same perception system, were deployed to operate cooperatively in real-field conditions. MLSysOps enabled dynamic coordination between the two platforms, allowing the drone to reinforce the tractor’s perception when detection quality degraded and ensuring reliable, targeted spraying throughout the operation.

Final field evaluations demonstrate the impact of ML-driven, predictive decision-making at the edge. MLSysOps agents continuously analyzed telemetry from the tractor and drone to anticipate upcoming degradation in weed-detection performance and proactively engage the drone before performance dropped.

The MLSysOps Smart City application demonstrates intelligent energy management for urban smart lampposts equipped with cameras and noise sensors. Deployed across real-world testbeds in Aveiro, Portugal, the system uses ML models to predict traffic activity from noise patterns, enabling proactive activation and deactivation of energy-intensive computer vision components.

The framework achieved significant energy savings while maintaining detection accuracy: in controlled, low-traffic environments, energy consumption was reduced by up to ~21%, while maintaining a 98.8% vehicle capture rate.

In high-traffic settings, the system prioritized reliability, achieving >99% capture rates with minimal energy savings (~3.5%), demonstrating adaptive behavior based on operational context. The ML-driven policy successfully balances sustainability objectives with safety-critical performance requirements across heterogeneous urban deployment scenarios.

Building on our strong technical results, the Smart Agriculture use case of the MLSysOps project was disseminated beyond the core consortium to engage a broader technical and industrial audience.

Final results and system-level insights of the Smart Agriculture use case were shared at the HiPEAC event in Krakow, Poland. This dissemination effort was followed by a visit to CNH’s Warsaw offices, where the project results were presented and discussed with their engineering team, fostering technical exchange and valuable feedback from an industrial perspective.

The MLSysOps project recently hosted its flagship Hackathon, “SysOps in Action”, bringing together over 60 students to explore the frontier of autonomous cloud-edge computing.

The participants formed seven teams and worked under the guidance of experienced mentors from across Europe to solve the practical hurdles of designing and applying different policies for cloud-edge continuum computing, with the primary objective of optimizing application performance while reducing operational costs.

At a glance:

• The Talent: 60+ students joined by expert European mentors
• The Structure: 7 teams working in a fast-paced environment
• The Challenge: Balancing application performance with resource efficiency
• The Tech: MLSysOps Framework, built on Kubernetes, Karmada, Grafana and OpenTelemetry
• The Outcome: Seven innovative solutions demonstrating how diverse perspectives fuel technical breakthroughs.

Watch our Hackathon video on [youtube] or read more on [our website]

MLSysOps successfully co-organized the 2nd Workshop on Machine Learning for Edge-Cloud Systems (ML4ECS), held on January 26, 2026, in Krakow, Poland, as part of the HiPEAC 2026 conference. Jointly organized with the Horizon Europe projects CODECO and EDGELESS, the event brought together researchers to address the complexities of ML-based autonomic management in the IoT-Edge-Cloud continuum. Profs. Nikolaos Bellas and Raffaele Gravina from MLSysOps co-organized the event.

Dr. Lorenzo Valerio (IIT-CNR) gave the keynote address on “Decentralized Federated Learning Over Edge Networks,” setting the stage for discussions on Agentic AI and coordination-free learning substrates.

The dedicated MLSysOps technical session featured a wide range of presentations demonstrating the project’s technical details, outcomes and impact across multiple domains. Following an architecture overview by Prof. Spyros Lalis (University of Thessaly), partners presented cutting-edge solutions, including:

• Smart Cities & Precision Agriculture: Real-world use cases were showcased by Ubiwhere on smart city analytics and by CNH on drone-tractor collaboration for targeted weed spraying.
• System Optimization & MLOps: Technical deep dives included UCD’s work on AutoML pipelines, Chocolate Cloud’s traffic-aware object storage, and hardware acceleration techniques presented by University of Thessaly (FPGA-based ML inference) and NUBIS (vAccel for hardware acceleration).