AURA: AI Research Tool for Journalists
Learn how a cross-border collaboration between the Indian Express, DR, The Economist, and Aftonbladet led to the development of an AI tool to help journalists enhance their research abilities.
[IMAGE CREDIT: Yasmin Dwiputri & Data Hazards Project / Better Images of AI / AI across industries. / CC-BY 4.0]
Journalists spend hours piecing together information, tracking down sources, verifying facts, and making sense of complex data. It's a process that takes time, often more than the fast-paced news cycle allows. With AI becoming more advanced, could it help journalists work through research more efficiently while keeping human judgment at the core?
That question brought together a team of journalists, technologists, and data scientists to develop AURA, an AI-powered tool designed to assist with research, helping journalists structure and analyse large amounts of information. Developed through the JournalismAI 2024 Fellowship, AURA isn't about replacing journalists but finding ways AI can enhance their work.
The team behind AURA included Vivek Arya, Chief Product Manager, and Krishan Kumar Arya, Chief Manager of AI/ML Tech from Indian Express in India; Tabtim Duenger, Data Scientist, and Alok Jha, Science and Technology Editor from The Economist in the UK; Aida Kokanovic, Digital Project Manager and OSINT Journalist from Danish Broadcasting Corporation (DR) in Denmark; and Martin Schori, Director of AI & Innovation from Aftonbladet in Sweden.
A Cross-Border Collaboration
For many of the team, the appeal of the JournalismAI Fellowship lay in the opportunity to collaborate across borders. "I wanted to create something together with other people," explains Schori. "I found the idea of collaborating with people from all over the world very appealing."
Duenger said, "It was the exact intersection of things that interested me and the opportunity to do something a bit different from my day job. Being early in my career, I wanted to network, meet more people, and discover different perspectives on how people approach journalism in this industry”.
“I was particularly interested in working towards using AI within journalism. There's been a massive emphasis on doing that internally where I work, but this project was unique. I'm in data science, so I work with technology a lot of the time. But building an actual generative AI or something with an LLM tool was super interesting to me,” she explained.
For Arya and Kumar Arya, the Fellowship allowed them to better understand AI's role in journalism.
"We wanted to understand how AI can help in journalism, as we are not from the journalism space. He's from the tech, and I'm from the product," said Arya. He explained, "AI was just starting to grow in terms of the use cases that people were building in journalism. And still, it's in a very nascent stage".
He added that they wanted to go beyond the basic applications they had read about. "When you read articles based on the use cases, they are very generic, but we thought that maybe we'd get some great ideas to work on and not just summarisations and basic functionality."
Arya said the Fellowship also allowed them to connect with professionals they wouldn't normally work with. "It helped us in reaching out to other people, interacting with people from both editorial and tech from other teams, because we had never interacted with any editorial person apart from our country," Arya said. "That broadened our horizons on how people look at things, how AI can help people."
Kokanovic valued collaboration: "I've always been excited to collaborate on projects. I think that we can sit alone with our projects and focus on one track, but when we are collaborating, things tend to expand."
For Jha, it was about improving journalism processes: "I've always been interested in looking for ways to improve the process of journalism," he said. He added, "I thought that the Fellowship would be a good way of meeting people to see how the sausage is made how you build a product."
Building AURA: From Ideas to Tool
Team AURA at the JournalismAI Fellowship meetup in London (Nov. 2024)
"We set the idea at the first meeting, and then we just stuck with it," said Schori.
The team combined their ideas into a product. "When we brainstormed on our topics, some of the key pieces got knit together, and we wanted to build something that aligns with each of us,” said Arya. Explaining that "it was a mix or amalgamation of ideas from everybody. We just built it from the top-down approach, we just scaled it. And so our initial ideas were similar, and we created a unique piece out of it."
What emerged was a tool focused on helping journalists structure unstructured data. Kokanovic described their goal as "How to get information from unstructured data". She explained how they identified processes that could be automated: "The research process is extremely time-consuming. With scientific papers being published daily in huge volumes, selection becomes challenging. Our approach allows you to input everything and then extract different perspectives."
Jha said he was gratified by the collaborative process of software development, particularly the partnership between editorial and technical teams. "I discovered how much editorial input matters in building tools. Initially, I wasn't sure what I could contribute as a non-programmer, but I found I could help by providing feedback.”
He added, "Rather than having someone just build something for you and you try it out, it was good to be able to, from the ground up, say look, I'd like it to look like this or I'd like it to be able to do that.”
Jha emphasised the importance of building a practical tool for journalists: "If it's not practical for me as a reporter to use, then I don't see the point in building it.”
The team incorporated Stanford University's STORM (Synthesis of Topic Outlines through Retrieval and Multi-perspective Question Asking), an AI-driven research tool designed to automate the creation of comprehensive, Wikipedia-style reports on various topics. STORM operates by gathering information from multiple sources, generating outlines, and producing well-structured articles with citations.
Arya said the opportunity to collaborate and learn from Stanford was particularly valuable. "It brought our project visibility to integrate STORM into AURA," he said.
Skills and Growth
Kokanovic gained technical skills through her work on AURA. "I gained so many technical skills. I am building another system right now that is similar to what I always wanted to build. If I hadn’t been on this Fellowship, it would’ve taken me a longer time to learn what I have, and I would’ve needed a course at the university because this pushed me to develop my technical skills in a way that I didn't imagine at all."
She also developed project management skills: "That was something that I didn't expect I would get, but when there are many team members with different time zones, we don't see each other, and we all want to contribute, then how do you land on something? Those were some of the skills that I also gained and quite frankly quite useful skills."
Kokanovic said the Fellowship "was beneficial, and sometimes even more challenging because we had to push ourselves in ways we probably wouldn't have to in our usual work environments where systems are already in place.” She said she is interested in continuing to pursue global collaborations and cross-border projects.
For Duenger, the Fellowship improved her technical and soft skills: "It helped my communication skills, being the most junior of the team and developing some confidence in my skills and speaking up in conversations."
She described her technical growth as, "Looking at front-end engineering and the User Interface of the tool, I've not done anything like that before. And that's something I take into my job now, so I can build streaming apps easily now as front-ends for my models to show internal stakeholders."
Duenger also valued learning about AI agents: "Our tool used 11 agents to orchestrate everything. I'm grateful I had the chance to work with and create some sort of agent orchestration pipeline."
Arya Kumar said, "During this journey, I gained insights into AI implementation. I learned about the best LLMs in the industry and their pros and cons. I also learned how to build a RAG model over LLMs, understanding how to utilise databases, embeddings, vectors, and create a product that could help journalists and editors."
Jha said he gained an understanding of AI development: "Understanding how these tools are built and how long it takes and the kinds of process you require to sort of do it properly—checking things, making sure that results are verified, how you collect data—that was all very interesting."
AURA in Newsrooms
Each organisation represented in the team is exploring ways to implement AURA within their newsrooms.
At The Economist in the UK, Duenger explained they are "in the process of trying to deploy this just on a server somewhere internally for a few of our editorial members to have a look at."
Jha elaborated: "We're going to host it, tweak it, freshen it up a little bit, and then demonstrate it in a couple of weeks. The idea then would be to pick under 15 reporters here, ask them to use it for a couple of weeks, and give us their feedback."
Arya and Kumar Arya are planning a demonstration for their editorial team at Indian Express. Kumar Arya explained that they want to refine AURA to work with regional languages such as Marathi, Hindi, and Bengali to see how AURA might perform with those regional languages.
Kokanovic said the DR plans to adjust AURA to fit their specific requirements: "We are thinking of adjusting AURA a little bit. We need to because AURA is good to go if we are just sending it out to everybody and every journalist out there. It's very much good to go, but we need to make it a little bit more organisation-specific because all the data used was trained on The Economist. We also need to make the output language Danish, of course."
She explained how they would tailor it further: "In DR we want to, for example, if we have one of these storylines, we want to diversify it in these directions that we work on, for example 'enlighten me,' 'educate me,' and all of these kinds of things."
Kokanovic interestingly shared that her newsroom also plans on integrating a tool from another team that participated in the Fellowship. However, it is still in the early stages.
Jha believes that AURA, once refined to work as intended, could offer more value than other deep research tools because it is specifically designed for journalists' needs rather than just producing extensive reports. He explained, "What AURA does, first and foremost, is decide whether something is newsworthy or not based on your input, which none of the other ones do."
Jha added that AURA can be tailored to specific editorial needs: "You can train it to what your news values, and when you put a new piece of information in, is it something you should cover." He elaborated that AURA "gives you a multi-agent search of the internet that will then produce a Wikipedia page for you about a topic."
Jha noted that while other research tools are "much more polished" with "clearly much more readable and well organised" results, AURA's journalistic focus gives it unique advantages that "the other tools just don't have."
Lessons for Cross-Border Collaboration
When asked what advice they would give to others considering international collaboration on AI journalism projects, the team said:
"People from different countries and ethnicities have their approaches, use cases to solve, ideas and ways of thinking, " said Arya. He suggests people should be "open to ideas, interpretations, and advice”.
Duenger echoed this, saying: "Listen to all of your colleagues because the different perspectives are useful. You'll always find something new you haven’t thought of from a different angle. And lots of people will have experience with different use cases as well. So take those learnings and try to work around them to implement them in the tool you are building”. Adding that communication was key.
Kokanovic advised, "Stay true to your passion because this journey can be a bumpy ride. There will be moments when you'll want to give up entirely. But knowing that you are working on a project you're truly passionate about, that's a huge win."
Arya also emphasised the importance of passion. “We wanted to do something passionately; it was not a side project for us, and everybody on our team was self-motivated, working to their full potential,” he said.
Despite the big difference in time zones, with India being 5 hours ahead of Europe, Arya said that their passion and commitment meant that geographical and time differences simply didn’t matter.
As Kumar Arya reflected, "It was a fantastic journey, and I never thought I'd be part of something like this. It was a great learning experience, and this journey will help me in every aspect of my career."
For Arya, the impact achieved in a short time made the experience valuable: "It was a short and sweet project, but it was remarkably impactful. Creating something in such a brief period that generates this much interest, where people see your work and immediately ask if it's live and available to use, that kind of platform is something only the Fellowship could have provided."
Schori added, "I would encourage people to do this [Fellowship] because you will get the chance to network and work with people from other countries and cultures. It broadens your perspectives."
Lasting Connections and Future Collaboration
For many in the team, the relationships formed during the Fellowship became more important than the product they built.
Schori reflected: "When I think back at it, I think it's a lot more about the people than maybe the product we created."
Arya agreed: "The relationship we built during this project tenure will outlive the product that we built."
The highlight for many was the in-person meeting after the Fellowship. After nine months of virtual collaboration, meeting face-to-face solidified their connections. At the end of the Fellowship, all the participating teams met up in London. "It was a great moment when we've been sitting on Google Meet for seven months and then met in person. That was a highlight," said Schori.
Jha said, "I loved meeting them and thought they were smart. I wish we'd had more time together, maybe a couple of weekends or a workshop in one place."
Kokanovic and Arya both echoed similar sentiments about their in-person meeting, with Kokonovic describing it as "the icing on the top”. Whilst Arya said it is "the finishing of the cake,” noting that "bonding virtually and then bonding personally, these are all different things."
For Team AURA, the Fellowship experience has led to lasting friendships. At the end of April, Kokanovic will attend a wedding in India and has decided to extend her trip to sightsee the rest of the country with Arya and Kumar Arya.
Watch Team AURA’s presentation at the 2024 JournalismAI Festival