ETH Graph Search: navigating the information jungle at ETH with AI
«ETH Graph Search» – an innovative search approach through the information sources [jungle?] of ETH Zurich
«ETH Graph Search» began in mid-2020, when Paul Cross from Institutional Research (IR), ETH Vice President Finance and Controlling (VPFC), asked ITS SIS for support on whether, and how, the ETH search could be improved with AI. The motivation behind it:
- It is often difficult to find ETH relevant information via ETH Search, and the ETH Search functionality often does not provide satisfactory results
- Francisco Pinto and his team from the «Centre for digital education» at the École Polytechnique Fédérale de Lausanne (EPFL) have done a lot of R&D (Research and Development) on the topic and developed the so-called «EPFL Graph Search» (the template for «ETH Graph Search»)
Connecting knowledge intelligently
This was followed in 2020 by a workshop on «Next Generation ETH Search Using AI» with contributions from IR, HK, ITS SWS, ITS SIS, EPFL and others for ITS AL (management) and members of the Digital Transformation Committee (DTC). A small proof of concept was financed in 2021. However, it took a great deal of lobbying before ETH was able to push through the development of its own solution. In 2023, the time had come and ITS SIS was commissioned to develop a prototype for an independent «ETH Explorable Graph» modelled on the EPFL solution (later renamed «ETH Graph Search» in line with «EPFL Graph Search»). Due to the interest, the prototype was then developed into a stable product in 2024 and into a service in 2025 with the first official release to a selected user group. In November 2025, there was a second release for all ETH Zurich employees.
Knowledge graph instead of search chaos
For the general motivations, functionality and main target group of ETH Graph Search, please refer to the ETH internal news article.
Stephan Zimmermann from IR/VPFC has been responsible for general management and product ownership since 2024. ITS SIS has been responsible for the technical management and overall development of the system from the outset. This includes the choice of suitable software components (e.g. libraries, databases), system architecture, deployment, etc. The preliminary R&D work of the EPFL team with their system as a blueprint, and their generous support of our endeavour with knowledge, experience and computer codes, helped us enormously, especially on a conceptual level. It quickly became apparent though that the different environments of the various information sources and IT infrastructures on which «ETH and EPFL Graph Search» are based also required different technical implementations.
Schematic representation of the architecture and choice of components of the «ETH Graph Search»
What good searches really need: Graph Search and data quality
ETH Graph Search is an undertaking with many exciting challenges at various levels (technical, organisational, etc.), from which we can also learn for future projects – e.g. for the «ETH AI Platform». Perhaps the biggest challenge of the «ETH Graph Search» is to keep the data quality of the tool consistently high. ETH Graph Search only provides added value to its users if the data displayed is up-to-date and correct. This means that we must constantly adapt to the dynamics and changes in the various information sources (e.g. APIs (Application Programming Interface). And that is indeed often a big challenge at ETH :-)
A big «thank you»
Tarun Chadha, ITS SIS (technical management); Vinzenz Muser, ITS SIS (overall development / deployment); Janis Fluri, ITS SIS (development); Richard Wartenburger, ITS SIS (deployment support); Silvio Mombelli, ITS APPS (provision of and support with importing OIS data); Adrian Fischli, Daniel Locher, ITS INFRA (k8s service and deployment support); Francisco Pinto and the team from the «Centre for digital education», EPFL.
Text & Contact
Dr Thomas Wüst, Group Head Computational and Data Science Support (ITS SIS), IT Services
erstellt am
in News
