Media Party New York @ Brown - 2026

Independent newsrooms face a structural trap: the functions that sustain journalism - audience development, grant management, impact measurement, revenue operations - compete with journalism itself for budget and capacity. AI is changing that equation. This talk is an honest account of what that looks like in practice, and the potential for newsrooms to stop choosing between the two.

Code for Africa, the Pulitzer Center and Earth Genome have partnered to build Africa Mining Watch, a satellite imagery and machine learning platform that detects mining operations across the continent, adapted from Amazon Mining Watch. This talk walks through how the three organisations are structuring transnational investigative journalism projects around the tool, pairing continental data leads with targeted one-on-one mentorship for African journalists. It also covers the community-building model behind the platform, including a mapathon bringing together journalists from across Africa to interrogate the data and develop original story pitches.

BOTO is an AI-powered WhatsApp bot created by InfoAmazonia to help people find trusted, relevant, and easy-to-understand news about the Amazon on the platform they already use every day. Built with large language models, BOTO responds to a simple but urgent problem: in much of the Brazilian Amazon, WhatsApp is a primary digital channel, while access to reliable journalism remains uneven and fragmented.
The project uses AI to deliver geolocated news, explain complex issues in simple language, summarize stories, personalize updates based on user interests, and collect audience input for future coverage. But BOTO is not limited to InfoAmazonia’s own reporting. It is being designed to include stories from other outlets selected through InfoAmazonia’s editorial curation process, based on reliability and relevance. The goal is to create a trusted entry point where users can find credible news about the places and issues they care about, regardless of which outlet produced it. Rather than asking people to navigate brands, websites, or search results, BOTO brings together trustworthy journalism in one conversational space. The product is being developed through an audience-centered process that includes testing, focus groups, and continuous feedback from real users across Amazon states.
In this keynote, I’ll share the core question behind BOTO: how do you distribute trustworthy journalism in places where platform habits, connectivity constraints, and local information needs do not match the assumptions of mainstream digital news products? Then I’ll show how we are using AI inside WhatsApp not as a gimmick, but as a practical layer for discovery, distribution, and engagement in public-interest journalism. Finally, I’ll reflect on what we are learning about accessibility, editorial trust, audience behavior, and the role of messaging platforms in the post-search era of news.

"Local Models to search through (Epstein) files" will cover tool building and data analysis methods using local models, with Epstein files as the case study, but applicable to any big data set a newsroom might have. The purpose is leave the workshop with the tools to go through a simple set up of local models and an llm-based analysis that can power an internal tool interface that can be vibe-coded. All these tools, techniques and frameworks can later be applied to the prototyping process that teams will need to go through for the hackathon part of the event. 

At Alley, we're working on an open-source audience toolkit for publishers. We observe that mid-sized and larger media organizations have fewer and worse options for payment acceptance, customer data and identity management, and authentication than do startup and indie publishers. We launched our tools on LAist.com, a major public radio digital platform in Southern California, and are soon to launch them with Chicago Public Media. I'll show our work, describe the project, and discuss the mandate to a new era of open source technology in the LLM era of news product development.

At NowThis, we don't have a big engineering team. So I built AI tools myself to solve the problems our newsroom actually has.
In 5 minutes, I'll walk through three internal tools I shipped:- A synthetic audience panel: AI-generated Gen Z personas that simulate focus group feedback on content before we publish, helping us catch tone and framing issues early.- An AI ad ops assistant: a Slack bot that manages paid campaigns across Meta and TikTok, pulls real-time analytics from API and data warehouse, and saves our team hours of manual reporting.- An AI pilot tool: lets our team generate AI-powered pre-pilots to visualize new show concepts before committing resources
I'll share what actually worked, what flopped, and the practical lessons for any lean media team thinking about building (not buying) AI tools. No massive budget required.

Small, mission-driven newsrooms face a persistent editorial bottleneck: too many pitches, too few editors, and a set of editorial values around representation, framing, and sourcing that resist easy automation. At Minority Africa, a Uganda based publication covering minorities on the continent, we wanted to know whether AI could help triage our now too many pitches without flattening the judgment behind it.
In 2025, we built Iraka, a custom GPT-based tool trained on a structured dataset of over 400 de-identified historical pitches paired with editorial decisions, reviewer comments, and outcome labels. The system prompt encodes Minority Africa's editorial criteria (including originality, representational depth, feasibility, and alignment with the publication's justice-oriented mission) and the tool returns a decision prediction (accept, develop, or reject), a score out of five, and written feedback designed to mirror the developmental tone of our editorial process.
In phase one, we benchmarked Iraka against real pitches it had not previously seen. It matched editorial decisions about 70% of the time, but feedback alignment was significantly lower at around 30%, showing that while decision replication is achievable, generating the kind of nuanced, contextual feedback editors actually give remains a major gap. We also tried to build it across different models with varied outcomes.
Phase two, now underway, targets the feedback layer directly. We are building synthetic and counterfactual pitch sets — controlled variations where elements like framing, sourcing, or representation are intentionally altered — to stress-test the tool on edge cases and ethical dilemmas absent from our historical data.
This talk shares what we built, what it got right, where it broke down, and the harder question underneath: what does it mean to encode editorial values into AI when those values center justice and representation?

The Global Index of Responsible AI measures how countries are honoring their international commitments on AI governance and responsible AI, especially those tied to UNESCO’s Recommendation on the Ethics of Artificial Intelligence. The project tracks 17 responsible AI topics, assessing whether countries have relevant laws, regulations, and real-world implementation, using a global network of 135 researchers—one in each country—who contribute data every two years. Beyond serving as a diagnostic tool for international organizations, the project also has strong value for journalists because it offers open, comparable data that helps contextualize public debates on AI, identify regulatory gaps, and bring a globally informed but locally grounded perspective led by a South African organization.

Stop chasing the latest viral distractions and consider the insights from three real-world case studies that demonstrate how building engaged audiences with journalistic impact can lead to long-term success. SembraMedia’s co-founder Janine Warner draws on research on more than 3,000 media organizations and news creators in the Project Oasis global media initiative to share a few innovative gems, and one cautionary tale.

What happens when you embed an AI engineer inside a newsroom and tell them to help wherever needed? At ProPublica, it led to “Spreadsheet Inference,” a method that combines the rigor of spreadsheet reporting with the scale of AI. This lightning talk will give a brief overview of how real investigations became product iterations, not the other way around. You'll see how the team turned iteration and empty into documentation, then code notebooks and eventually a fully featured Google Sheets plugin.