NEWS Organisations today struggle to handle vast amounts of information stored across multiple databases, registers, and systems. In his doctoral thesis, Romuald Esdras Wandji at ͯÑÕÊÓƵ presents a new solution that may transform how both humans and AI systems work with data – in a safer and more accessible way.
Research by Dr. Romuald Esdras Wandji at the Department of Computing Science, Umeå University, opens the door to faster and more secure data handling across complex systems.
ImageHans KarlssonRomuald Esdras Wandji at the Department of Computing Science has developed a new framework that enables not only the retrieval but also the updating of data through Virtual Knowledge Graphs. The technology has existed for decades, but until now, it has only worked in one direction – users could query information through a conceptual layer, while any changes still required technical expertise and direct manipulation of the underlying databases.
This new research makes knowledge graphs bidirectional, allowing both humans and AI agents to correct, add, and update information through the same conceptual layer they use to search for it.
“My research helps remove a major bottleneck by enabling both people and AI assistants to manage data using familiar concepts rather than technical languages,” says Romuald Esdras Wandji. “What previously required specialist developers can now be carried out by anyone. It fundamentally changes who can work with data.”
Most organisations store their information in many different systems, each with its own technical structure. Virtual Knowledge Graphs have made it possible to retrieve data from these systems through a shared conceptual model – but only for reading. Wandji’s research addresses this long‑standing limitation by giving the conceptual layer update capabilities. This allows users to work with a single, unified view of information — both when retrieving data and when modifying it.
Data is one of society’s most valuable resources, yet managing it still requires specialised technical skills. This creates bottlenecks, slows down decision‑making, and increases the risk of errors.
“By allowing both humans and AI agents to work through a clear and semantically structured layer, these barriers are reduced. It becomes easier to correct mistakes, add new information, and keep systems up to date – without writing database code,” says Wandji.
The technology is particularly relevant for organisations that handle large and complex datasets, such as public authorities with extensive records, healthcare providers integrating patient information across systems, research‑intensive fields like biomedicine, and large enterprises with fragmented data environments.
“Complex information systems can become more flexible, more accessible to domain experts, and far more adaptable as organisational needs change,” Wandji adds.
As AI systems take on greater responsibility for interacting with data in both the public and private sectors, secure data management becomes increasingly important. Giving AI direct access to databases carries significant risks.
With Wandji’s framework, the knowledge graph acts as a safety layer: AI systems interact with data through a controlled conceptual vocabulary, reducing the likelihood of incorrect or harmful updates.
“As AI systems assume more responsibility, it becomes crucial that they operate through structured and understandable interfaces. My research contributes to that foundation,” says Wandji.
Romuald Esdras Wandji defended his thesis at the Department of Computing Science in May 2026. Principal supervisor: Professor and world-leading expert in artificial intelligence for data management, Diego Calvanese, , (The Wallenberg AI, Autonomous Systems and Software Program). Download the thesis "Ontology-based Update in Virtual Knowledge Graphs"
