Projects

HES-SO

• 2025 – 2029 DEEP: Deep Learning Resource-Efficient GPU Orchestrator. A project funded by the Swiss National Science Foundation (SNSF).
• 2025 – 2026 OptiCloud: Optimisation de l’utilisation et de l’impact du cloud. An inter-domain project funded by HES-SO. In collaboration with Exoscale.
• 2023 – present AI center for SMEs (french-speaking Switzerland). Funded as a flagship project from HES-SO.
• 2023 Energy EmoMaps. Inter-disciplinary project funded by Axe Transformation Numerique, HES-SO Valais-Wallis. Online platform here.
• 2022 – 2024 Anomaly detection and data pipelines. Collaboration between HES-SO Valais-Wallis and the Lonza Group.

Swiss Young Academies

• 2024 The Spark podcast. Stories from Advocates for Global Change. https://en.swissyoungacademy.ch/joint-projects/the-spark-podcast
  
• 2024 UPSTEM: Unveiling the leaky pipeline in STEM. https://en.swissyoungacademy.ch/joint-projects/unveiling-the-leaky-pipeline
• 2024 AI in science and society. https://en.swissyoungacademy.ch/joint-projects/ai-in-science-and-society
  

SDSC

• 2020 EasyFAIR. Swissuniversities. In collaboration with ETHZ, ZHAW, HSLU.
• 2019 Renku proposal for GCP research credits. Google Cloud Platform GCP research credits grant.
• 2021 Renku Integration at FSO. Mandate with the Federal Statistics Office.
• 2020 BAG Health Data Science. Collaboration with the COVID Task Force.
• 2020 Deploying Renku in an offline environment. Collaboration with the Health2030 Genome Center.
• 2019 Renku deployment at Openshift. Mandate and collaboration with Red Hat.

EPFL

• 2013 – 2018 Towards Resource Efficient Datacenters. Grant from Microsoft Research. Swiss Joint Research Center.
• 2013 Kairos preemptive data center scheduler.

Kairos is a preemptive data center scheduler presented at SoCC 2018. that runs without the need of task runtime estimates. It is implemented as part of Yarn and it uses Docker containers to achieve a lightweight preemption. Source code.

• 2015-2017 Eagle and Hawk hybrid data center schedulers.

Hawk is a Hybrid Data Center Scheduler presented at Usenix ATC 2015. It takes the best of both worlds combining centralized and distributed schedulers. It has the following main features:

1. 1. Hybrid Scheduling. Schedules Long jobs in a centralized way (better scheduling decisions) and Short jobs in a distributed way (better scheduling latency).
   2. Work stealing. To do better load balance when a node is free it will contact another one and ‘steal’ the short-latency-sensitive jobs in the queue.
   3. Partitioning. It prevents Long jobs from taking all the resources in the cluster so that Short jobs do not experience head-of-line blocking.

Eagle is a hybrid scheduler which aims to avoid the Head-of-Line blocking that short jobs experience in distributed schedulers by providing and approximate/fast view of the long jobs. Source code.

• 2013 Systems and Networking Simulation Research for Fabric Computers. Research internship at Microsoft Research Cambridge, research group lead by Anthony Rowston.
• 2012 Large-scale task scheduling in Hadoop. LABOS research group lead by Willy Zwaenepoel at EPFL.
• 2012 Domain Specific Language for Distributed Algorithms in Scala. Master final project at LSR group lead by André Schiper at EPFL.
• 2011 Invariant Verifier for Parallel Programs Master semester project at LAMP group lead by Martin Odersky at EPFL.
• 2011 STAMP in Java: Benchmark for Software Transactional Memory. Master semester project at LPD group lead by Rachid Guerraoui at EPFL.