Generating a meal plan from a prompt is one well-crafted LLM API call — that part is genuinely easy. Turning it into a shopping list with accurate nutrition data is the part that needs a real, licensed dataset behind it.
Borderline
Effort: a weekendRunning cost: $0/mo to start
Why CanICodeThis says Borderline
Asking an LLM for a week of meals matching dietary preferences and returning structured output (JSON) is a well-documented prompt-engineering pattern
Saving favorite plans and regenerating individual days is ordinary CRUD on top of the same API call
A shopping list generated by combining ingredients across the week is straightforward text processing
The parts that'll cause problems
Accurate nutrition data (calories, macros) needs a real food database — a free public one like USDA FoodData Central covers this without needing to build your own catalog
Recipes the LLM invents on the fly can be inconsistent in quality; grounding generation against a curated recipe set produces more reliable results
Dietary restriction handling (allergies, specific diets) needs careful prompt constraints to be trustworthy, not just a checkbox
How I'd build this
frontend
Next.js
database
Supabase
auth
Supabase Auth
hosting
Vercel / Cloudflare Pages
MVP scope
A form for dietary preferences and days needed
One LLM call generating a structured week of meals
A combined shopping list from the week's ingredients
Save and revisit past plans
Postpone to v2
Nutrition data pulled from a real food database rather than LLM estimates
Regenerate a single day without redoing the whole week
Grocery delivery integration
What it actually costs to build
AI agent — with a subscription (Claude Pro/Max, Cursor, etc.)
$0 marginal
AI agent — pay-per-use API, no subscription
$43–$86 one-time
Hosting, once it's running
$0/mo (free tier)
Domain name, if you want your own
~$12/yr
The build prompt
Build me an AI meal planner, genuinely achievable with careful data sourcing:
1. A form for dietary preferences (vegetarian, allergies) and number of days needed.
2. One LLM API call returning a structured JSON week of meals with ingredients — ask for JSON explicitly, don't parse free text.
3. A page combining all ingredients into a single shopping list grouped by category.
4. Save generated plans to revisit later.
Use the USDA FoodData Central API for nutrition estimates rather than trusting the LLM's own numbers — that's a real, free, authoritative data source, not something to approximate. Out of scope: regenerating a single day without redoing the whole week, which needs the plan to be edited in place rather than freshly generated, and grocery delivery integration, a real, separate per-provider integration.