Consumer AI · Agentic App · Solo Build
Plum Kitchen: Dinner, decided
- Role
- Sole builder, idea to production
- Stack
- Lovable · Supabase · Claude API
- Outcome
- Live and in daily use · three weeks of real usage cost $0.67 total
The problem
I looked at what already existed, Mealime, Paprika, before building anything. Both are well-built, and both share the same assumption: that you already know what you want to cook and just need help organizing it. That's not my problem. I get inspired constantly, I'll spend twenty minutes saving recipes and thinking about ingredients, and then by the time it's actually time to cook, I'm exhausted, the inspiration's gone, and I default to the same three meals. The gap isn't organization. It's the transition from inspiration to committed action, and nothing I found closed it.
What I built
Plum takes whatever you give it, ingredients, a recipe link, a half-formed craving, and reasons across your goals, constraints, and cooking history to plan your week and build the grocery list. No organizing required from you. Plum runs on a set of LLM calls that each do a different job: interpreting vague input into structured recipes, reasoning across your stored goals, constraints, and history to plan the week, and reconciling every ingredient across those meals into one grocery list. The MVP does four things. It generates full recipes from vague input, so "that tahini noodle dish I saw" becomes something you can actually cook. It builds ingredient cards that reason about how to use something based on your goals, not a generic database lookup. It plans the week by weighing everything at once: what you've saved, what you're aiming for, what you've already eaten, what you're tired of. And it generates the grocery list from the finished plan, reconciled and deduplicated across every meal. I built it solo on Lovable and Supabase with the Claude API, and that stack was a deliberate choice. My value isn't writing backend infrastructure, it's knowing what to build and why, and this let me move fast enough to test the product thinking instead of getting stuck in engineering.
The result
Plum is live, and my husband and I have been the test users. We recently moved in together, we both love food, and the meals we eat together matter to us. Plum is what makes that happen more often than it otherwise would. It isn't perfect. Latency shows up at the moment you least want to wait, and I'm not convinced the planning output reflects my preferences as well as it should yet, which tells me the feedback loop needs to be more visible, not just smarter. It's ripe for another turn based on what real usage has taught us. But it's been a shining light of hope on a Tuesday night when we're tempted to order Goop Kitchen.