bebi

I built bebi for a problem I was living with every day as a parent. Planning balanced, age-appropriate meals for a young child is harder than it should be. The tools available are either adult meal planners with child features bolted on, generic recipe collections, or trackers that create more work for already tired parents.

The gap I wanted to close: a focused, parent-first meal planning experience for babies and toddlers.

A mobile-first meal planning app for babies 6–24 months. Core flows: weekly meal planning calendar, food logging, recipe browser, family sharing, and achievement-based engagement loops. Built using Lovable and Capacitor as a React app that behaves like a native app on iOS and Android. Shipped to closed beta with five users from my discovery interviews.

Decision 1: Position as a meal planner, not a food tracker. Most existing competitors focus on tracking — what did your baby eat today, allergic reactions, nutritional logs. I made the call to position bebi as a planner, because the real pain point parents articulated in discovery wasn't "I forgot what I fed her last Tuesday." It was "I don't know what to make tomorrow." Tracking puts the cognitive load on the parent: decide → log → review. Planning puts the decision step inside the product: suggest → plan → grocery → log. That framing shaped every downstream decision, and simplified the solution across two distinct users — the parent who knows what they want to make but needs a planning tool, and the parent who needs a lot of guidance on what foods to serve.

Decision 2: Pivoted the build approach mid-project after AI tools failed to recreate the designs faithfully. I started by building mockups in Figma using Figma Make, then tried to build bebi using Lovable, Cursor, and Replit. All three struggled to faithfully recreate the designs. Replit completely failed and Lovable was closer but still hallucinated constantly and would change numerous things even if I just requested a small change. I lost weeks fighting this before stopping to research alternatives and talk to other builders. The fix: pivot to a React app wrapped in Capacitor instead of native React Native — a build path I could actually iterate on with AI tooling while still shipping something that behaved native. For the few things that needed native capability, like the push notification strategy, I could build just those components in React Native. The lesson I carried forward: instead of asking AI tools to build UI from visual designs, I now export component files from Figma directly, upload those to the build tool, and use the AI to build the API services around working components — much faster and cleaner, because I'm giving the AI what it can work with and recognizing each tool's strengths versus what they aren't.

Decision 3: Deferred AI meal suggestions for v1. The obvious "AI-native" feature would have been LLM-generated meal plans personalized to each baby. I deliberately deferred this for two reasons. First, AI generation in a regulated category — pediatric nutrition, allergens, safety — compounds liability significantly. Second, the actual user need in discovery could be solved by curated recipes from trusted sources, so AI-generated meal plans would have been overengineering a v1 in terms of cost, complexity, and risk. AI stayed on the roadmap as a v2 enhancement after validating the core thesis with lower-risk infrastructure.

Decision 4: Paused production after consulting a lawyer about strict liability in the category. This was the hardest call. The category sits at the intersection of pediatric health claims, food allergen liability, and parental reliance — strict liability territory. Even a perfectly executed product carries personal exposure for a solo founder if a child has an adverse reaction, even if it has nothing to do with your app. The compliance infrastructure needed — allergen disclosure systems, medical advisory review, claims substantiation, COPPA-compliant data handling — was a six-figure investment before I would earn a single dollar.

I paused. Not because the gap isn't real — it is, and I still strongly believe this needs to be built — but because solving it responsibly requires a company with legal, compliance, and trust-and-safety functions. Solo founder isn't the right vehicle for this category.

Most PM case studies are about shipping. The pause decision on bebi was the most difficult PM decision I made last year because I stepped back and, despite being in love with the product, knew it wasn't the right move forward to ship it. Knowing when to stop — and being able to articulate exactly why — is harder than knowing when or what to ship sometimes, especially with sunk-cost bias at play.