Lume

My co-founder and I built Lume between Nov 2025 and Jan 2026 (3 months to launch), at Station F in Paris. We're both ex-gaming addicts, and we wanted to build the thing we wished existed when we were struggling.

I handled the backend architecture, real-time systems, AI translation pipeline, and overall technical decisions. We both coded the frontend together with Claude Code. We shipped on iOS and Android in 28 languages.

There are countless habit trackers and recovery tools for standard addictions, but nothing specifically made for gaming addiction.

From a product perspective, this created a unique UX paradox: we needed to provide digital tools for people fighting screen addiction, without making our app their new addiction. We had to design something that offered support and replaced the gaming habit with real-world activities, deliberately avoiding the dark patterns and gamified dopamine loops that our users were trying to escape.

Instead of another habit tracker, we built an empathetic, intervention-based tool. We worked with researchers specialized in Internet & Gaming Disorder (IGD) to ground everything in validated science.

• Emergency "craving" interventions with quick-action buttons and a 24/7 AI companion for high-stress moments
• A learning center born from our conversations with addiction researchers, helping users understand their brain chemistry
• Real-world hobby suggestions based on the user's budget and preferences
• An anonymous community chat with live AI-powered translation so users globally could support each other in their native language
• Sobriety tracking, daily reflections, progress dashboards

To maximize reach, we leveraged AI to natively localize the entire app into 28 languages. Not just translation, we made sure the content sounded natural to local users. We also engineered the live-translation feature in the community chat so people from different countries could talk to each other seamlessly.

We designed, developed, and deployed to both app stores in just 3 months. Got 300 active users globally, largely driven by the SEO benefits of our deep 28-language localization strategy. Received really positive feedback from users who were grateful the tool existed, though the deeply personal nature of addiction made it hard to extract actionable feature feedback from the broader base.

We officially stopped active development in January 2026 (the app is still live). We hit a wall with monetization. Asking people to pay for addiction recovery feels wrong, and users felt that too. The deeply addicted ones don't even know they need help, and the ones who do feel uneasy being charged for it.

We realized building the kind of trusted, engaged community needed for this to work would take years. The motivation was 100% personal for us, but we didn't have the runway.

Validation over building. Before writing a line of code, we should have spent time in deep conversations on forums and communities. Build the audience first, then build the product for them.

The power of localization. Shipping a high-quality product in 28 languages makes users in smaller markets incredibly grateful. Technically, this also pays massive dividends in SEO and organic traffic.

AI-driven velocity. This project proved we could design, build, and ship a complex, localized, multi-platform product incredibly fast with just two people, by treating AI (Claude Code) as a true extension of our engineering workflow. That execution speed carried into everything we did after.