A journalism game about truth, speed and people.
Built in two days at Breda University of Applied Sciences for the KreativEU Hackathon, by a team of four pulled together from across Europe: myself, André Costa, Federico Ruffini, and Vladimir Khvostuntsev.
What KreativEU Is #
KreativEU, short for Knowledge & Creativity European University, is a European University Alliance bringing together higher education institutions from across the continent, including Breda University of Applied Sciences. The alliance connects local and national heritage, traditions, crafts and folklore with a shared set of European values, with the aim of building a genuinely European education and research system where students, researchers and academics collaborate across cultures, borders and disciplines. Hackathons like this one are one of the ways that vision plays out in practice, bringing students from different partner universities together to build something in just a couple of days.
The Challenge: Values in Play #
This edition, hosted in Breda, was called Values in Play, and the brief was to build an AI augmented game that puts European values at the centre of the gameplay itself rather than bolting them on as a message at the end. My team and I decided to build a game around journalism, since few professions sit as directly at the intersection of truth, speed, public trust and the real consequences of words as reporting does.
What the Game Is About #
The Midnight Chronicle puts the player in the seat of a reporter at the local newspaper of a small town. Information keeps arriving in news cycles: police statements, emails, phone calls, posts online. For each one, the player judges how reliable it is, then chooses to publish it, check it first, or hold it back. If they publish, they write the headline themselves. Readers react to exactly what was written, and every choice shifts how many people read the paper, how much they care about the story, how much they still trust the player, and how much media pressure the reporting puts on the real people involved.
Who It’s For #
It’s designed as a learning tool for students from middle school to high school, aged 14 to 18, the age when opinions about the news start to form. Teachers can run it in media literacy or citizenship lessons as a single 20 to 30 minute session, followed by a class discussion. Every story in the game is stored as a plain JSON case file, so teachers aren’t limited to the cases we built. They can write their own, closer to their own classroom, their own students, or a situation relevant to their community, without touching a line of code.
Heritage, Responsible AI and Sustainability #
Local news is a community’s shared memory. It records a town’s people, events, traditions and dialect, and it only works when people trust it. We wanted The Midnight Chronicle to bring that piece of heritage into the classroom, letting students experience first hand how careful reporting keeps a community’s trust, and how misinformation can break it.
Responsible AI was a core design principle throughout, not an afterthought. The AI only ever gives advice and voices readers’ reactions, it never decides what’s true, and the player always makes the final call. Every use of AI is clearly disclosed to the player, and the readers’ comments carry built in safeguards against insults, threats, slurs, and invented facts.
The game is also built to be sustainable in a very practical sense. It runs fully offline on an ordinary laptop, with no subscriptions or API costs, which makes it easy to actually use in a real classroom rather than something that only works in a demo. It’s also straightforward to integrate into different environments beyond the one we built it for.
My Role #
My own work on the team sat across AI engineering and frontend design and functionality. On the AI side, that meant building out the AI editor’s tone and consistency checks with the NLI model, and wiring up the local LLM for reader reactions, structured outputs, safety constraints and all. On the frontend, I worked on the look and feel of the game itself, from the hand-drawn SVG interface to how the player actually moves through a case, judges a source, writes a headline, and sees the consequences play out.
Features #
Under the Hood #
The game runs fully offline on an ordinary laptop. The backend is written in Python, served by a lightweight HTTP server built on the standard library, and holds a rule-based game engine handling source judgements, publishing and fact-checking, trust and interest scoring, media pressure per character, competitor scoops, corrections, multi-case progress, and end-of-case timelines.
Two local AI models sit on top of that engine. The first is an AI editor built with Hugging Face Transformers using facebook/bart-large-mnli, doing zero-shot tone classification and natural language inference to check whether the player’s own notes support or contradict their headline, combined with rule-based detection of dramatic wording, named people, and anonymous sources. The second is a small local LLM, Llama 3.2 3B running through Ollama, which generates readers’ reactions to each published headline using structured JSON output, safety constraints baked into the prompt, and pre-written fallbacks when the model isn’t available. A caching layer stores AI results so rehearsed demos load instantly.
The frontend is a single-page application in plain HTML, CSS and JavaScript, no framework, built around a hand-drawn SVG visual style: a title screen, a case board of dossier folders, an office desk with a clipboard of incoming sources, a printed-newspaper view, a stepped green-to-red pressure scale, an agenda-style notebook, and SVG line charts for players to analyse the impact of their own decisions. Every story is stored as a JSON case file, so teachers can write their own cases without touching code.
Responsible AI was a core design principle throughout: the AI only ever advises and never decides outcomes, every use of AI is disclosed to the player, all scoring stays transparent and rule-based, and no data ever leaves the device.
How It Went #
Two days is a brutal amount of time to design a game engine, train and wire up two separate AI models, and build an entire hand-drawn interface from scratch, let alone do it as a team that had never worked together before this hackathon. By the end of day one we had a rough skeleton. By the end of day two, we had a working case, a working AI editor, and reader reactions actually generating in near real time.
We ended up taking first place at the hackathon. More than the result itself, what stuck with me was how naturally the idea of “values in play” fit a story about journalism. Every mechanic we built, judging a source, weighing a headline, watching trust rise or fall, ended up being a small lesson in its own right, which is exactly what a media literacy tool should be.
The Midnight Chronicle is still a hackathon prototype, built in two days rather than two months, but it’s one I’m genuinely proud to have in my portfolio, both for what it taught me technically and for what it reminded me about why responsible, human centred AI design matters.