Beyond Chatbots: Novia Develops Interactive AI Workspace for Engineers
In the photo: Christoffer Björkskog and Lamin Jatta at a conference in Koszalin, Poland.
Over the past months, researchers at Novia University of Applied Sciences' Maritime Technology Research Group, Christoffer Björkskog and Lamin Jatta, have been developing something that pushes AI interaction far beyond the traditional chat window.
Novia's own platform, ANCHOR, developed within the Virtual Sea Trial project, facilitates collaborative work with an AI agent using a dynamic visual canvas. The platform allows users to explore documents, run simulations, delegate and review work, and build knowledge structures in a way that is intuitive, transparent, and genuinely collaborative.
Instead of hiding its reasoning behind text in a chat, ANCHOR visually displays how information is connected. Nodes, links, sources, simulations, and results appear directly on the canvas. Users can upload PDFs, FMUs, and other data sources, and the AI agent can extract parameters, run simulations, interpret results, and organize everything into a living, interactive workspace.
“In a nutshell, ANCHOR is a showcase for the recent transition towards agent-native apps. These center on a collaborative workflow in which humans and agents alike use the same tools to work together in a shared workspace. In ANCHOR’s case, the canvas is the shared workspace, and the agent can, for example, be Claude Code, Codex, or Open Code. The logic being that you bring your own agent into the app to collaborate on the shared canvas”, says Team Leader Johan Westö.
“Agents are becoming real users of our software. We should design for them as first-class citizens, not bolt them on afterwards”, says Christoffer Björkskog.
Why Are You Developing This?
"We are developing this because technical documents contain a great deal of valuable engineering knowledge, but that knowledge is often difficult to extract, verify, and reuse. Engineers frequently need to search manually through PDFs, datasheets, manuals, and simulation files. This process is both time-consuming and prone to errors," says Lamin Jatta.
The goal is to create an AI-assisted workspace where knowledge from technical documents is not only accessible through conversation, but also visualized, structured, and linked directly to supporting evidence.
What Will It Be Used For?
ANCHOR is being developed to support engineering knowledge work involving technical documentation and simulations.
"For example, it can extract facts, specifications, tables, and visual regions from technical documents. It can also show where each answer originates in the source material. The platform organizes findings on a visual canvas and connects extracted document parameters to simulation workflows. In the future, it will also support FMU-based simulation analysis," explains Christoffer Björkskog.
Who Will Use It in the Future?
The primary users will be engineers, researchers, students, and other domain experts who work with technical documentation and simulation models.
"This includes marine and mechanical engineers, simulation engineers, and researchers working with digital twins or virtual commissioning. Students learning how to connect technical documentation with simulation models can also benefit from the tool. Product development and maintenance teams that need to understand technical specifications quickly are another potential user group," says Jatta.
How Does This Differ from Current Methods in Virtual Simulation?
"Current virtual simulation workflows often require users to manually read documents, extract parameters, interpret tables, and transfer values into simulation tools. The connection between the original document source and the simulation input is often weak or lost entirely. This project is different because it aims to keep the entire chain visible, from the source document and extracted evidence, to the structured canvas, and finally to the simulation parameters," Jatta explains.
This means users can see not only the simulation inputs, but also where those inputs originated and whether they are grounded in the original documentation.
How Are You Developing It Further?
"We are continuing to improve the document medallion pipeline, consisting of a bronze layer for raw documents, a silver layer for structured pages, text, tables, and metadata, and a gold layer for semantic regions, screenshots, and source-grounded evidence," says Jatta.
"The next steps are to make information retrieval more reliable, improve how the agent chooses between text, tables, lists, and images, and connect extracted parameters more directly to FMU-based simulations. Our long-term vision is an explainable engineering assistant that helps users move from technical documents to validated simulation workflows," Björkskog concludes.