Summer 2024 Fellow Reflections

Angela Voit is a rising senior at the University of Michigan with a bachelor’s in Data Science. She served as the Tribune’s Summer 2024 Engineering & AI/ML Fellow. Learn more about Angela here.

What was your path to the Tribune? Why did you want to apply?

My interest in data journalism and newsroom engineering was sparked a few years prior to joining the Tribune when I saw the dashboards, guides and live data reporting on COVID-19 and the elections in 2020. During such a confusing period, I saw many news organizations going beyond the traditional article by article style of reporting to find new and innovative ways of communicating truth. Soon after I joined my college paper, the Michigan Daily, to see how I could contribute.

After a couple years with the Michigan Daily, former Tribune Fellow Eric Lau dropped the AI/ML Engineering Fellowship info in our Slack channel. It seemed like the perfect opportunity to find out what it was like to work in a professional newsroom and explore the rapidly developing field of AI/ML with a commitment to accuracy, accountability and ethics that is supported throughout the organization.

What did you do during your fellowship? What have you learned?

I’ll highlight three projects from my fellowship.

HelpBot and AskArticles

These projects functioned at first as experiments to see what the Google platform we were using strived at and where it fell short. My first week project was building a chatbot off of our Content Management System’s FAQ page and eventually became an all around tool for technical assistance for the Tribune, HelpBot. AskArticles was a bot to ask questions about our article data. While they now use the AWS platform, experimenting with them helped me learn about the components of chatbot development.

The projects served to help me learn about retrieval augmented generation (RAG), chunking/tokens and prompt engineering. I also got to use Slack Apps for chatbot integration and Zapier for automatically tracking feedback. Moving to AWS, Suraj found semantic chunking and access to newer models which significantly improved the performance of the model.

Voter Question Search Guide (name in progress)

The primary overarching project of my fellowship was to explore the possibility of featuring a public-facing Question and Answer AI agent for our voter guide content. My primary contribution to this project was pivoting from a chat option to a search and generation option. The tool responds to voting questions with a list of our relevant voter guide content and then uses that content to generate a summarized response. We hope to publish it before the 2024 election to help Texans navigate the voting process.

On this project I explored the advantages and limitations of Google tools. This included both the backend of the search integration and the user interaction elements in their DialogFlow messenger and search features. The most significant discovery was in the improved clarity and accuracy of a Search-first RAG over the more open-ended chatbot design.

Trusting News Research

In partnership with the Trusting News organization, the Tribune conducted a user research project on the audience's thoughts, feelings and perspectives on AI and its use in the newsroom. I helped select interviews, conduct interviews, and analyze survey data. The project kept the desires of the audience front and center as I worked on our AI tooling.

Conducting a user research project really rounded out the skillset of my fellowship. I was able to utilize some data techniques for analysis and interviewee sampling while also practicing interviewing skills and turning our observations into next steps. In calls with Trusting News, I learned about the cutting edge of trust-building research. It hinged on a central concern of my project: how can we ensure that the use of AI in journalism doesn't further erode the audience's trust in the news?

What was the most surprising part of the fellowship?

I think number one was the cute kids (and occasionally cat) in stand-up. Alas, only every once and awhile but certainly a lovely surprise every time!

But as far as the work of the team, I was surprised by the product-focus. Every member of the team was mindful of the user experience of the product. It was great to work with an engineering team that was thoughtful of the end goal throughout the development process.

What is your favorite memory from the fellowship?

One of my favorite memories of the fellowship was when my mentor Chris connected me with the data journalist Yuriko to help out on her heat deaths analysis project. Not only was it exciting to be able to bring in techniques I had been learning in school to a project with real impact, but I loved to meet researchers advising the project and get a feel for the editorial side of the Tribune.

What is your advice for anyone applying?

Don’t worry if you don’t think you have enough experience to apply. What matters in these 10 weeks is not the experience you bring in but your openness to learning as you go.

Sometimes what a new hire can do best is not write the most code but instead bring a fresh perspective to the project through the learning process. My experience was more centered around learning and testing unfamiliar out-of-the-box tools over coding, so while I had some web dev experience at the start, it certainly wasn’t core to success.

So if you’re curious, apply! You might be surprised in the future that awaits you.