MLSpace – a step towards crucial infrastructure for the Machine Learning/AI era

Mapped principles 

The project is mapped to the following framework principles:

P5. Available experimental environments

P3. Active student-centred learning 

P4. Assessment and examination for learning  

Definitions of the principles

Contact

Jayanth Raghothama associate professor

jayanthr@kth.se

Profile

Project number 

This project is implemented at the School of Engineering Sciences in Chemistry, Biotechnology and Health (CBH) within the Future Education programme (project no. 2613-CBH) 

Project context 

The cultural shift towards professional model development lifecycles is critical in the Machine Learning (ML) /Artificial Intelligence (AI) era. A first step needs to be taken towards building the infrastructure that can enable this transition. The outcome of such a project would benefit not only education but can also be used by researchers. 

DevOs is a cultural, technical, and process-driven approach combining software development (Dev) and IT Operations (Ops) to shorten development lifecycles and deliver high-quality, reliable applications rapidly. Similarly, MLOps (Machine Learning Operations) is a set of practices, workflows, and tools that unify Machine Learning (ML), application development (Dev) with IT Operations (Ops) to automate and streamline the lifecycle of ML models. It bridges the gap between data science and production, enabling faster, more reliable, and scalable deployment, monitoring, and retraining of models. It eliminates silos, fostering automation, improved collaboration, lifecycle management and automation, and reliability and transparency in machine learning model development.  

Many students now work on ML, either in courses or increasingly for their theses. However, these efforts remain as independent projects and assignments, spend a lot of (manual) effort configuring experiments, tracking and comparing results, versioning data etc. In many of these courses and theses, students rely on Jupyter notebooks, work on curated datasets, and focus primarily on modelling methodologies and metrics. From a pedagogical point of view, the integration of MLOps has the potential to transform “how to train models” into “how to build models for the real world”. From a research point of view, MLOps improves reproducibility, transparency, and enables collaboration across student groups. There is however, no open-source framework or infrastructure that is developed to facilitate these requirements in an educational context.

Purpose (outcome) 

The purpose of the project is to take the first step towards building the infrastructure necessary for professional model development lifecycles, which is critical in the Machine Learning (ML) /Artificial Intelligence (AI) era, by developing software to enable the integration of MLOps in courses, theses and research. 

Outcome goals (examples): 

• Students will spend less (manual) effort configuring experiments, tracking and comparing results, versioning data etc and have more time to further advance their projects. 
• Students’ learning transforms from “how to train models” into “how to build models for the real world”. 
• Researchers working on ML will benefit from a professional and systematic way of running ML experiments, which improves reproducibility, transparency, and enables collaboration. 
• Increased shareability: While the deployment will be connected to KTH Cloud, the software will be shared within KTH to anyone interested, to be deployed on their own computational infrastructure.  

Further characteristics and benefits of the intended software are described in the project directive (see link below). 

Project results (output) 

The project will deliver the following concrete results: 

• A software published as a shared resource across KTH, for anyone to deploy and use, that can:  
• Create model registries, share datasets and enable students to post model scores and explainability metrics as results.  
• Publish student model scores as a leaderboard to be visible in courses. 
• Split training and test sets and publish only training sets.  
• A pilot test of this software in one course (Applied Machine Learning and Artificial Intelligence, CM2011).  
• Example scripts and demonstrations on the usage, for students, educators and researchers.  
• A report on the lessons learnt, further improvements.  

The development will follow an agile project management methodology.  

Time plan 

Start date: 2026-05-01
End date: 2027-04-30 

Project documentation 

If you have a KTH ID, you can read the documents when logged in: 

• Poster: The project’s first poster (in Swedish) will be available at the ”Storträffen” meetup autumn 2026
• Project directive (Swedish): Projektdirektiv_FrU26_2613-CBH_MLSpace.pdf