Case Study

Lenfest AI Fellow spotlight: Peter Rasmussen

The Baltimore Banner

October 19, 2026

This profile is part of the Lenfest AI Collaborative and Fellowship Program’s 2026 Impact Report.

As an AI Fellow at The Baltimore Banner, Peter Rasmussen is building tools that directly support both impactful journalism and the Banner’s business operations. We spoke with Peter about the lessons he’s learned so far, what excites him most about AI, and where he believes the industry is headed next.

About his role and the problem he is trying to solve

Peter Rasmussen is the principal AI engineer at The Baltimore Banner. His work focuses on building practical tools that help journalists go faster and deeper, and on helping the business solve real problems.

His flagship project is News Detector, which has several applications.

The first is to scrape news from publications across the internet and assign them a newsworthiness score based on criteria suggested by editors. The goal: To save editors from having to manually scan several different news sites every morning. To do this, Peter developed a system where the news is provided to editors, with a layer of news judgment already applied on top.

News Detector also contains a tool that lets journalists use natural language to query what the Banner and other publications have written about a particular topic, for research. The tool was partly inspired by Lenfest AI Fellow Kevin Hoffman and the Philadelphia Inquirer team’s projects, he said.

“It’s nice having that collaboration because that made things for us easier, as they shared their code with us.”

Also useful: A “foodie bot” that scrapes Instagram posts and turns them into a weekly digest, integrated into Slack so an editor can more easily scale a concept that was originally vibe-coded but too expensive to run at scale.

Peter is also working on a tool to sift through audience tips, separating genuine leads from self-promotional pitches, and to power an events calendar that surfaces “top things to do” in a user‑friendly way.

Peter came to AI in local news with two degrees in engineering and a career that led him through environmental consulting in New Orleans and a year of work in Kenya, where he worked for a startup focused on energy and agricultural initiatives. He also worked as a management consultant in Washington, D.C., where he helped build an infrastructure mapping and navigation platform, a project that sparked his interest in data science.

He then completed a master’s degree in AI at Johns Hopkins University, focusing on computer vision and geospatial analytics, including projects that used satellite imagery to detect heavy construction and other changes.

He was a Banner subscriber when he saw the job posting, “and I was like, ‘Oh my god, I gotta apply to this.’” The chance to combine his technical skills with mission‑driven local journalism was too compelling to pass up.

The biggest surprise during the fellowship

One of Peter’s biggest ongoing lessons is how hard it is to estimate effort accurately. For example, he thought he would be able to finish building the News Detector scraper in a week, over his Christmas holidays.

“It took five weeks. I wasn’t done until the end of January,” he said.

Another major surprise has been just how powerful agentic coding can be. Early on, faced with a long list of requirements for the fellowship and for his role, he realized he couldn’t just hand‑code everything himself. Instead, he focused on building a harness to govern the behavior of LLMs, creating a structure that constrains and regulates the agents.

The combination of better frameworks and rapidly improving models has delivered a “step‑change in productivity.”

One thing the industry gets wrong about AI

Peter is clear that he does not see AI as a replacement for journalists and believes some organizations misuse AI by treating it as a substitute for human reporting.

“We have human beings who are experts – that’s the truly differentiating factor.”

At the Banner, he sees a healthier approach: Use AI as an assistant for summarizing, surfacing information from outside the newsroom, and understanding customer behavior, while keeping editorial judgment firmly human.

“Everything else has to be vetted by humans,” he said, pointing to the rigorous daily editorial conversations he overhears from his desk.

Lessons that other newsrooms could apply tomorrow

Peter offers two tightly linked lessons: build scalable frameworks and start from real use cases.

“Build frameworks that allow you to scale your productivity,” he said. His agent harness has now been adopted by other engineers, including colleagues at the recently acquired Pittsburgh Post‑Gazette. One of their engineers, he notes, now “has a better setup than I do,” and is tackling a massive migration project with impressive productivity.

The second lesson is about requirements discipline: “Really understand what the use case is, what the needs are, and elicit the use cases, because otherwise you’re building something that isn’t going to be used.”

In other words, don’t start with the model. Start with clearly defined newsroom and business problems.

What he is paying attention to right now

Right now, Peter is closely tracking rapid advances in large language models, both proprietary and open source.

What he is most optimistic about

Deepening the Banner’s understanding of its audiences and using that insight to build better products and more sustainable revenue.

His recent work clustering users by engagement and connecting that to donor propensity is already showing that “people who are more engaged tend to… donate more.” He is developing a blueprint for donor modeling that leverages those insights as well as learnings from Chicago Public Media AI Fellow Mark Chonofsky.

Peter is also excited about using AI to take repetitive, time‑consuming tasks off people’s plates, for example, building structured rubrics to help HR sift through hundreds of resumes for a single position, “basic but onerous tasks that take forever” and keep staff from doing higher‑impact work.

“I’m so optimistic that we’ll be able to take care of a lot of the stuff that’s just a pain… that people don’t like, that’s time-consuming and onerous, and we can take that off people’s plate.”

The Fast Four

A book, article, or newsletter you’d recommend?
Doppelganger by Naomi Klein.

What’s one thing you’re learning that has nothing to do with AI?
How to create a “lush garden,” digging up weeds and understanding wildflowers, in his front and back yards.

What local journalism challenge deserves more attention?
Resourcing and workload. “People wear multiple hats and are very busy. Workflow automation is crucial to relieving that pressure.”

One word that describes the next year of AI in news?
Adoption.