An AI debate app shipped solo to 1,000+ users

A 1v1 debate app. Two people argue one topic against a clock, and an agent pipeline fact-checks and scores each turn in under 15 seconds.

The idea

Online arguments reward whoever is loudest. Who's Right put a frame around them. Two people who disagree, one topic, timed turns, and an AI referee that scores how well each side argues instead of who talks the longest.

Three ways to duel

Live, turn by turn. Asynchronous, spread over a day. Or local, in the same room, with one phone on the table between the two of you. In local mode, speech-to-text splits the conversation into turns and each side gets a clock.

An iPhone lying on a table between two drinks, showing the local duel screen listening to a speaker. — The phone sits between the two speakers, splits what it hears into turns, and runs a chess clock for each side.

The judge

An orchestrator reads the message. A fact-checking step verifies the claims against live web sources. Then a panel of judges scores the turn on rigor, evidence, rebuttal, clarity and openness, which are the criteria used in competitive debate. I used Google Cloud Platform for the AI and n8n to orchestrate around 45 workflows on a Dockerized server. The pipeline takes under 15 seconds per exchange.

An n8n canvas showing an orchestrator node, a fact-checking node, a switch, and a fan-out of judge nodes. — An orchestrator node, then a fact-check against live web search, then the judges that score the turn.

What you get back

A verdict that answers two separate questions, who was right and who argued best. Then an ELO change, and a coaching report that quotes your own sentences back and points out the biases, the fallacies and the sources that do not hold up.

Shipping it solo

I built the app, the backend, the database, the UX/UI, the agent pipeline, the brand, the landing page and the go-to-market on my own. I had no coding experience before this. I worked with AI from the first line to the App Store release, and agents automated 80%+ of the operations. The project was incubated at Station F in Paris.

Why it is paused

I shipped five major releases and reached 1,000+ downloads. Paying subscribers came, but infrastructure costs stayed above revenue. I paused the app in April 2026 because the running costs were too high.

What I learned

Shipping a product alone from start to finish taught me more than any course. Next time I will check willingness to pay before building the full pipeline, and the infrastructure bill will be in the business plan from day one.