Hi there,

I’m a Product Manager and journalist-turned-technologist building AI-driven storytelling products at The Washington Post.

⇩

My Work

The world’s first interactive news podcast

News consumers are reading fewer articles but listening to more podcasts than ever. The problem is, podcasts are expensive to make and can't really be personalized. So we set out to build one using AI that could: a podcast generated just for you, based on what you care about and how much time you've got. I worked closely with newsroom editors to get the tone right, then built the system that turns every published article into a fresh audio segment and stitches them together based on each listener's interests. We designed it to feel less like "talking to a chatbot" and more like two friends catching you up on the news.

  • We wanted this to feel like a utility, not a gimmick — something you'd reach for to catch up on your commute, not a novelty you try once. The goal was "hack your schedule," and the tone was two friends breaking down the news at whatever pace you actually have time for.

  • Getting an LLM to write news scripts that actually sound like good journalism took a lot of back-and-forth with editors — checking quality, adjusting the prompt, checking again. Once that was dialed in, every new article could generate its own script automatically. Then we pre-produced audio blocks and stitched them together on the fly based on what each listener cared about, so no two episodes sounded quite the same.

  • Users finished these episodes at a meaningfully higher rate than our regular podcasts. We also learned something surprising — more on that below.

  • Users are more wary of AI "personalities" than we expected. There's real distrust baked in right now within the market. That's pushing version 2 in a more no-frills direction. Less emphasis on personality, more on utility features that genuinely help users stay informed.

Building a news homepage for the AI era

Fewer people are clicking into articles these days — they're getting their news secondhand through social feeds and AI summaries instead. That makes the homepage one of the last places a news org still fully controls. I'm leading the research and strategy behind a redesign meant to balance "here's what matters" with "here's what matters to you" — backed by deep audience interviews and a teardown of 30 competitor sites. One thing jumped out of the data that we weren't expecting: our most loyal, every-day readers were churning faster than our casual ones. That's flipping the whole strategy — instead of one homepage for everyone, we're building different experiences for different types of users.

I led a competitive analysis that calculated attributes of every major competitor so designers were able to work from a data-informed POV. I used AI tools to complete a task within hours that would have taken days.

The analysis determined we were an outlier in many areas such as content density and the ratio of headlines-t0-visuals.

We also uncovered a problem behind the scenes that impacted the UX: editors were wrestling with 200+ settings just to lay out a single story. So I designed a curation system that builds story packages automatically, freeing editors up to focus on judgment calls instead of manual layout work.

I conceived and designed Stacker, a new internal curation system that improved the user experience by standardizing the design. It also helped editors work faster in breaking news situations.

  • A big chunk of our audience checks the homepage multiple times a day. That habit felt like the thing worth protecting and rewarding, so I built a personalization approach around keeping things fresh for repeat visitors instead of showing everyone the same static page. Clicking through to full articles is becoming rarer across the industry, so the bet was to pack real value right into the homepage itself, while still making it easy to go deeper for people who want to. And not everyone wants the same thing from a homepage — our most loyal readers want the latest, fast; casual readers want a quick, curated overview. One page, several different jobs.

  • We noticed a big red flag in the data: rising churn among our most loyal readers was the signal that changed our whole approach. From there, my team launched in-depth surveys with current and potential users to stress-test our assumptions. We also analyzed 30 competitor homepages to see where we stood — turned out we were heavier on density and lighter on visuals than most.

  • Editors were stuck using a tool with 200+ configuration options per story — powerful in theory, but slow and inconsistent in practice, especially during breaking news when every minute counts. I designed a replacement that automatically pulls together story packages — video, photo galleries, related content — so editors spend less time fighting a clunky tool and more time on the calls only they can make.

  • Early iterations have driven a small but real lift in same-day return visits, and cut scroll fatigue by 40%. The full redesign is in progress now, built on what we've learned so far.

Creating an evaluation system for AI-generated content

I built a fully interactive internal tool editors could use to evaluate the quality of an AI summary so we could fully judge the quality of the prompt. Editors could click on the information icon to see the description of the quality metric, which was provided by a managing editor.

Turning articles into AI audio briefings meant writing a prompt that could produce accurate summaries that sounded good to the ear. The process had previously been highly subjective and took multiple rounds of labor-intensive testing.

Meanwhile, internal and external trust in AI-generated content was shaky, which made getting this right — and being able to prove it — absolutely essential for a go-to-market strategy to succeed.

So I designed and ran the whole evaluation process. I worked with senior leadership to define what "good" actually meant — what criteria mattered, and how much weight each should carry. Then I built two things with Claude Code: a lightweight tool that let editors test a single piece of AI output at a time, without ever exposing the full prompt or proprietary content, and a dashboard where leaders could see results graphically and react directly to individual pieces of feedback.

  • “Is this good quality?” isn't something you can check with a simple pass/fail test. It needs to pass multiple granular attributes to reach a premium quality threshold. On top of that, this was happening against a backdrop of real public skepticism toward AI in journalism — readers and editors alike were primed to distrust it. That raised the bar: it wasn't enough for the prompt to be good, we needed a defensible, repeatable way to show it was good.

  • I worked with senior leadership to determine what attributes editors should grade (examples: accuracy, tone, whether it stayed true to the source). That turned a subjective evaluation process that involved dozens of editors into a structured, scalable workflow.

  • Given the internal backlash to AI, it was important to protect our exploration process. All it takes to sink an otherwise solid go-to-market strategy is an anonymous leak showing one out-of-context moment in the early evaluation process. The tool I build in half a day ensured all internal stakeholders only had one piece of the puzzle when evaluating their examples.

  • The leadership-facing side wasn't just a static report. Leaders could see results rolled up graphically, but they could also respond directly to individual pieces of editor feedback, which turned the whole thing into an actual conversation instead of a one-way status update. That's what made it a real human-in-the-loop system: every prompt version got real human judgment, from both editors and leadership, before the prompt was released in production.

  • I used Claude Code to build the whole system myself rather than filing a ticket and waiting on an engineering team's roadmap. It took half a day, from first idea to releasing a tool editors could use.

A pitch deck I created for The Washington Post’s inaugural WP Incubator program, where I was a finalized for this venture that turned archival IP into compelling storytelling.

MY WORLDVIEW

How I think about AI

I'll say it plainly: I'm bullish on AI.

The way we consume information is fundamentally broken. Articles are losing relevance, and even when people read them, they rarely walk away connected to the bigger picture — just another headline, disconnected from the last one.

I believe AI has a real shot at fixing that: making information radically more accessible, and helping people actually make sense of the world they're living in. That's the future I want to help build, and it's why I'm so drawn to opportunities to shape what news becomes next.

But I'm not naive about where we are right now. The AI backlash is real, and it's earned in plenty of cases. A good chunk of my work is making sure AI-generated content is actually accurate, and finding ways to build trust with news consumers who have every reason to be skeptical.

I think artificial intelligence right now is similar to the early internet: a tool that's going to be used for both real good and real harm. I’m interested in doing work that tilts that balance toward the good — accesible information that actually helps people understand the world.

Awards & Recognitions

    • Made the final round for The Washington Post’s inaugural incubator program cohort.

    • Conceived a venture that transformed historic news IP into rich media.

    • Won The Washington Post’s company-wide hackathon contest for using AI to transform article content into engaging new storytelling experiences.

    • Recognized as part of the team that upgraded the For You tab in The Post’s app, delivering more personalized recommendations through a refreshed user interface.

    • Led the creation and release of audio article playlists in The Post’s app, expanding our audio offering and improving engagement.

    • Led the team that redesigned the app homepage interface, launching a card-based experience that increased content discovery and user engagement.

    • Contributed to the development of a full-screen vertical video player, driving millions of video starts and modernizing the in-app video experience.

    • Led the redesign of the My Post tab into a premium, utility-driven space that inspired a broader app overhaul.

    • Was selected as a finalist after creating a niche news product focused on the intersection of politics and the Catholic Church.

My Resume

Download