It's 6:47 PM on a Tuesday. My kid is hungry. My wife is on a call. The fridge has: two eggs, a half onion, some limp spinach, a tub of paneer that expired maybe yesterday, and a bag of rice that's been open since Diwali.
I open DoorDash. I know I'm not going to find anything under $35 for the three of us. I do it anyway.
This happened roughly four times a week for about two years.
The problem isn't what you think
People assume meal planning fails because we're lazy. We're not lazy. We're decision-fatigued.
By the time dinner rolls around, you've made 200 micro-decisions at work, answered 40 Slack messages, remembered two appointments, and negotiated screen time with a four-year-old. Asking your brain to also creatively synthesize a nutritionally balanced meal from the random contents of your fridge is asking too much.
So we don't. We order. We guilt-eat. We repeat.
I tried every meal-planning app. All of them assumed I'd shop beforehand, follow curated recipes, and cook from an organized pantry. None of them answered the actual question:
What can I make right now with the weird stuff I already have?
The first version was a spreadsheet
I'm going to be honest. The first version of MealAI was a Google Sheet. Columns for ingredients, a formula that matched them to recipes I'd saved from the internet. It worked for me for about two weeks before I gave up maintaining it.
The second version was a Python script with hardcoded recipes. That lasted a month.
The third version — which I actually shipped — used Claude's API to take a list of ingredients and return three meal options. That was the moment it stopped feeling like a spreadsheet and started feeling like a tool.
What I learned building it
A few things surprised me:
Most "AI meal planners" aren't really using AI. They're pulling from a static recipe database and letting you filter. That works, but it fails the 6:47 PM test. If your fridge has paneer and spinach and you need dinner now, a database doesn't adapt — an LLM does.
Country matters more than cuisine. I'm Indian-Canadian. For a long time I thought "Indian food" was a useful filter. It's not. An Indian person in Mumbai eats differently than an Indian person in Toronto. What they have in their fridge is different. What they call dinner is different. The product had to detect country, not just ask about cuisine.
The hardest part is ingredient matching. If you say you have "rice," do you mean basmati, jasmine, brown, or the sad half-bag of Minute Rice at the back? Building a fuzzy matcher that handles "paneer" vs "cottage cheese" vs "Indian cheese" took me three weeks. I'm still fixing edge cases.
People don't want recipes. They want permission. The biggest unlock wasn't generating better recipes — it was telling users "you have enough to make dinner tonight, here's what." Turns out most of us have enough food. We just don't know it.
What I still haven't figured out
I'm writing this partly as a writeup and partly because I'm stuck on a few things and I'd love input from people who've built stuff in this space.
- Onboarding friction. Getting someone to tell me what's in their fridge is way harder than getting them to tell me their dietary preferences. I've tried photo-based scanning, a chat-style input, and a grid of common ingredients. The grid wins, but it still feels tedious.
- Trust. AI-generated recipes are great until the app suggests something with a texture mismatch (paneer with ice cream, once, don't ask). I now run every generated recipe through a sanity check before returning it. It's fragile.
- Business model. Free for the database-matched recipes. Pro for AI-generated ones. I think this is right but I'm not sure. AI calls cost money. Database matches don't. Feels fair.
Why I'm still building
I don't think MealAI is going to be a billion-dollar company. That's fine. It doesn't need to be.
What it has done is reduce DoorDash nights in my own house from four a week to about one. That's roughly $400 a month in our budget. It's also reduced the amount of food we throw away, which bothered me more than the money did.
If a few other families get the same outcome, that's a real product. It doesn't need to be bigger than that.
---
If you're the kind of person who also stares into a fridge at 6:47 PM wondering what the hell to make — I built this for you. Try it at usemealai.com. It's free. Tell me what breaks.
