Navigating rules and standards efficiently with DNV’s RuleAgent
RuleAgent, DNV’s latest AI tool, helps maritime professionals access DNV’s rules and standards faster and with greater clarity. What started as a lightbulb moment gradually developed into a core solution, benefiting from well-established company innovation processes.
When Erik Spieler joined DNV Maritime as a technical trainee in 2023, he expected a steep learning curve. What stood out quickly was how central rules were to almost every task.
“You very quickly have to start looking into the DNV rules,” he says, recalling his first months in technical support and shadowing surveyors at newbuild yards. “They form the basis for every answer we give.” But learning where to look proved harder than understanding the rules themselves.
The difficulty of navigating rules
Watching experienced colleagues recall precise rule paragraphs from memory, Erik saw how easily less experienced staff could end up on the wrong path. The real challenge was not interpreting the rules themselves but knowing where to look and understanding the context behind each requirement, from vessel type and class notation to detailed machinery configurations.
One moment in a shipyard stood out for him when, after hours of preparation, a senior surveyor told him his notes applied to the wrong ship type.
“That was when I thought there had to be a better way,” he says.
Applying large language models to maritime rules
Thinking more about the problem, Erik quickly understood that the complexity lay in the scale and detail of the rules, as opposed to the quality of the documentation. Classification rules, international regulations, and flag requirements overlap, and applicability depends on dozens of parameters, including vessel type, size, year of build, class notations, and much more.
With large language models (LLMs) entering the mainstream around the same time, Erik realized that their ability to read large volumes of text and understand context could be an ideal way to navigate the rules. He tested the idea with senior surveyors and technical experts and the response was unanimous.
“I realized that this could be the perfect use case,” he says. “Every surveyor I spoke to said the same thing: it would be a dream come true.”
RuleAgent concept moves from idea to venture board pitch at DNV
Following his lightbulb moment, Erik quickly set about translating it into reality.
DNV has a structured innovation framework with clear processes for helping employees turn novel ideas into action. Early in 2024, Erik was able to submit his initial concept through the internal innovation portal. This was well received, and Erik was soon after partnered with members from DNV’s Class Development management team.
“Over two days, we collaboratively developed the idea, benefitting from valuable input and strong support from the team,” he says. From there, he was invited to pitch to the Maritime Venture Board, a panel of senior leaders who, in “Shark Tank” style, would assess the concept, pose challenging questions, and decide whether to bring it forward to the next stage.
Turning early scepticism into a scalable concept
“There was positive feedback from the beginning, but also scepticism, which was very understandable,” says Spieler.
Two years ago, LLMs were still unproven in safety-critical contexts. Accuracy, traceability, and trust were open questions. Those questions defined the journey that followed.
As the concept matured, it also exceeded what one person could build alone, with diverse areas of expertise needed to take it to the next level. In mid-2024, Erik reached out to Martin Borge Heir who, at the time, was a digital trainee rotating in DNV’s AI Research Centre in China.
“It was good timing,” says Heir. “At that stage, it was still very much a research project. I instantly saw its potential and knew that I could support Erik to bring it to the next level.”
Understanding how rule searches affect real maritime operations
Before writing significant code, Erik and Martin had to validate the problem itself. They spent a lot of time interviewing internal technical experts and customers to understand how rule searches affect real operations.
The feedback was consistent. Time spent looking for the right rule or requirement delays decisions on board and ashore, disrupting operations, so any tools that could speed up operations would be widely welcomed. The feedback was used to move the concept further through DNV’s innovation framework.
“The next steps involved taking it through different stages and asking tough questions every step of the way,” says Heir. “The main principle is to stress test the concept at every gate and kill it fast if it doesn’t hold up. Most ideas don’t survive and that’s the point.”
Proving the technical feasibility of RuleAgent
Early work focused on technical feasibility. This meant testing whether an LLM could reliably surface the right rule paragraphs from a rule set spanning around 600,000 paragraphs across over 30,000 pages. That turned out to be the biggest challenge. Off‑the‑shelf approaches were not accurate enough, and the team quickly realized that they would have to build a system specific to the job at hand.
“If you just took an off-the-shelf AI model and used it on the rules, you wouldn’t get anything sensible out,” says Spieler. “The structure of DNV’s rules had to be understood and used deliberately, rather than treated like generic documents, and this meant that we had to build something bottom up, tailoring our development approach to DNV-specific needs.”
Designing for responsibility
The key pathways for development aligned around accuracy and relevance, two non‑negotiable design constraints. The team learned quickly that rules are never abstract. Applicability always starts with the ship, whereby its characteristics such as age, dimensions, class notations, flag, and history determine which requirements matter.
By combining vessel data with rule structure, RuleAgent could narrow its search far more effectively. Crucially, along with finding and summarizing the relevant rules for each case, RuleAgent was designed to also point users back to the original paragraphs.
“Verification remains essential and it’s vital that we can always show where each answer comes from,” says Heir.
From prototype to product
Throughout 2025, internal experts tested early versions and fed back what worked and – crucially – what did not. Prototypes were refined, discarded, and rebuilt.
“This was a pivotal stage of development,” says Heir. “We were getting constant feedback from colleagues, and this enabled us to make adjustments quite quickly.”
By June 2025, RuleAgent met the criteria for a minimum viable product. It was deployed internally in October, before being finally launched for external use in March 2026.
Widespread use of RuleAgent across maritime industry
Since its launch to the wider maritime industry, the uptake of RuleAgent has grown steadily. Engineers, surveyors, and customers have already become accustomed to its benefits, with feedback praising the way it helps them navigate rules more efficiently, particularly in time-pressured situations, such as inspections between ports.
“Applying AI to our rule framework supports more efficient decision-making and better navigation of an increasingly complex regulatory environment,” says Geir Dugstad, Technical Director, Classification, at DNV Maritime. “RuleAgent has already been proven to significantly speed up processes, from newbuild through to fleet in service, and this has a ripple effect on operations on the ground, providing real value to our customers.”
Expanding RuleAgent for further use
For Erik Spieler, the journey itself remains the most striking part. What began as a moment of confusion during his early career became a tool now used by highly experienced professionals, aided by a company willing to listen and with the right framework in place to make it a reality.
He maintains that these are still the early stages of RuleAgent’s journey. The longer-term goal is to bring all the relevant maritime rules and regulations into a single AI tool, with the DNV rules forming the first layer. Work is already underway to expand coverage to IMO conventions and flag-specific requirements.
“Applying AI to the rules and to our core business is really important and this is just the beginning for DNV,” concludes Spieler.
Get regular industry insights!
Join the thousands of industry leaders today who have already signed up to receive the latest insights.
sign up