Trinity Hall Fellow and Director of Studies in Physics and Astrophysics, Dr Boris Bolliet, in collaboration with fellow physics and deep learning researchers, has developed an AI-powered ‘scientific assistant’ designed to accelerate the scientific process. The AI agent can help scientists identify new research questions, analyse and interpret data, and produce scientific documents.
Meet Denario: An AI assistant for every step of the scientific process
Posted:
05 Nov 2025
Agentic AI on laptop screen.
The tool, called Denario, uses large language models to help scientists with tasks from developing new hypotheses to compiling manuscripts. The team hopes Denario will make research faster, more dynamic and more interdisciplinary.
AI can already help with parts of the scientific process: tools like ChatGPT can visualise data or write abstracts, for example. But these tools are typically limited to one step at a time. With Denario, however, scientists have developed a new kind of assistant: one that can synthesise existing papers, formulate new research questions, analyse data, and write manuscripts.
In a paper published on the arXiv.org preprint server, Dr Boris Bolliet, Dr Francisco Villaescusa-Navarro from the Flatiron Institute, and Dr Pablo Villanueva Domingo from the Autonomous University of Barcelona provide an overview of the tool and suggest that Denario could accelerate and broaden the scientific process, giving scientists the ability to use it for whichever aspect they find most helpful.
“Sometimes the most interesting thing is the idea, because maybe it’s a new idea that hasn’t been explored,” said Dr Francisco Villaescusa-Navarro, one of Denario’s primary developers. “Sometimes it’s a new method that’s never been applied to a certain dataset. There are many ways Denario can help expand the way we think and point us in new directions.”
The team stresses that Denario is not a replacement for scientists, however. The current version has major drawbacks: only about one in ten outputs yields interesting insights and, in some cases, Denario has fabricated data. Human review of Denario’s work remains essential.
Denario’s development was led by Dr Boris Bolliet, Dr Pablo Villanueva Domingo and Villaescusa-Navarro and included a team of researchers from astrophysics, biology, biophysics, chemistry, material science, neuroscience, mathematics, machine learning, quantum physics and philosophy.
With recent advances in large language models such as ChatGPT, Google Gemini and Anthropic’s Claude, the researchers saw an opportunity to test how such tools might perform across every stage of the research process.
Denario uses a collection of AI ‘agents,’ each specialising in a different task. While Denario can complete the entire research process end-to-end, the agents can also be used separately for specific steps.
To use Denario end-to-end, scientists upload a dataset along with a brief description of what they’d like it to do. The first pair of agents develops and refines ideas for how best to approach the dataset, generating potential research projects. The next set searches through existing research literature on the topic, assuring that the project idea is new and grounded in previous work.
Once the idea is refined, the methods and planner agents suggest approaches for analysing the data. The next agents follow through on these plans, using a multi-agent system called CMBAgent, which acts as Denario’s research analysis back end. These agents write, debug and run code, then interpret the results. Finally, the writing and reviewing modules produce and revise summaries of the findings.
“The agents all work together to make it possible,” Villanueva Domingo said, emphasising that scientists can easily check each module’s work and, if desired, run the agents individually.
So far, Denario has been tested end-to-end hundreds of times on datasets across disciplines including astrophysics, neuroscience, chemistry, biology and materials science. Most outputs were deemed unsuitable in expert reviews, but around 10% produced an interesting question or finding.
Because Denario can draw from multiple disciplines, the team is hopeful that it can identify new research questions that a specialist might never think to ask.
“Denario can pull ideas from other fields that maybe a scientist is less familiar with and would never have considered,” said Villanueva Domingo. “That interdisciplinary nature is very exciting.”
The researchers also hope Denario will help scientists win back some of their most valuable resource: time.
“I hope that Denario will help accelerate science by providing researchers with tools that allow them to spend less time on menial tasks — like scrolling the arXiv, formatting images, summarising analysis — and more time on deep creative thinking,” said Bolliet.
In its next iteration, the team aims to make Denario more efficient and capable of producing higher-quality work, including automatically identifying and filtering out low-quality outputs.
Tools like Denario still face challenges. Some of the system’s final write-ups didn’t adequately convey uncertainty in the results, and it sometimes struggled to reference previous studies clearly, even when it could describe their content accurately.
There are also technical and ethical considerations, including the risk of Denario drawing from AI ‘hallucinations’ — misleading or false information — as well as questions around copyright and authorship. The researchers even had to add an instruction telling Denario not to generate ‘dummy data’ after it produced fabricated results.
The team say they look forward to an open discussion about how best to use Denario and similar tools in the scientific process, as well as how to prevent potential misuse. They emphasise that Denario was only possible thanks to collaboration across academia and industry.
Originally published by the University of Cambridge on 5 November 2025.