AWS Builder Center
Weekend Annoying Task Challenge: Global Gluten Free Meal Planner

Weekend Annoying Task Challenge: Global Gluten Free Meal Planner

AI-powered weekly meal planner for gluten-free eaters who want culturally rich foods. Generates a 5-day dinner plan with per-meal gluten free (GF) safety notes and grocery list. Built for the AWS Builder Weekend Challenge. using S3, Lambda, Bedrock and DynamoDB.

Every week I sit down to plan meals and every week it takes longer than it should. As someone who eats strictly gluten-free and genuinely loves culturally rich food, I'm not just answering "what do I cook?". I'm also answering "is this safe?" and "do I have everything?" for every single dish.
I find that most meal planning tools either ignore gluten-free needs entirely or point you toward the same handful of recipes. And the gluten-free resources that do exist rarely reflect the depth and variety of cultural cuisines.
That's the annoying task I decided to automate: the weekly ritual of figuring out what to cook, verifying it's safe and building a grocery list from scratch.
Global Gluten Free Meal Planner is an AI-powered weekly meal planner that lets you select the cuisines you want to feature that week; it generates a complete 5-day dinner plan with per-meal gluten-free safety notes and a grocery list organized by supermarket aisle.
How I Built It I started with the user experience. My goal was a clean, minimal interface that felt warm and food-forward rather than clinical. The frontend is pure HTML, CSS and JavaScript with no frameworks or dependencies. Just a single index.html file hosted on Amazon S3 as a static website.
The core logic lives in an AWS Lambda function written in Python. When the user hits Generate, the frontend POSTs the selected cuisines, any special notes and the number of days to the Lambda Function URL. The Lambda constructs a detailed prompt and sends it to Amazon Bedrock using the Amazon Nova Lite model, which returns a structured JSON response containing the meals and grocery list. Each generated plan is also saved to Amazon DynamoDB, keyed by a unique plan ID.
The hardest part was prompt engineering.
Early versions of the prompt let the model do whatever it wanted. Select only one specific cuisine and it would come back with a variety of cultural dishes. I had to be very explicit: list the allowed cuisines, instruct the model to distribute meals evenly, require the grocery list to come exclusively from the ingredients in the generated meals and banned ingredients by name. Ingredients such as wheat, all-purpose flour, barley, rye, regular soy sauce with specific GF substitutes called out explicitly (rice flour, stone-ground cornmeal, GF tamari). Getting the model to stay within guardrails while still being creative took several iterations and it's where I spent most of my time.
The second challenge was CORS. The app is hosted on S3, which is a different origin than the Lambda Function URL. The Lambda was returning data correctly when tested from the command line but the browser was blocking every request. The fix was adding the correct response headers: Access-Control-Allow-Origin: *, Access-Control-Allow-Headers: Content-Type, and Access-Control-Allow-Methods: POST. Once those were in place everything worked immediately.
A smaller but visible issue: on regeneration, the old grocery list stayed visible while new data was loading, making it look like nothing had changed. Fixed by explicitly clearing both containers before showing the loading state.
AWS Services Used
Details
Amazon S3
Static website hosting for the frontend
AWS Lambda
Serverless Python backend triggered via Lambda Function URL
Amazon Bedrock (Nova Lite)
AI model that generates the meal plan and grocery list from a structured prompt
Amazon DynamoDB
Stores each generated meal plan for persistence
The architecture is entirely serverless. The user interacts with the S3-hosted frontend, which POSTs to the Lambda Function URL. Lambda calls Bedrock, parses the response, saves to DynamoDB and returns the data to the frontend for rendering.
What I Learned This challenge pushed me to think carefully about prompt design. The difference between a prompt that says "use these cuisines" and one that says "use ONLY these cuisines, no exceptions, every single meal must belong to one of the listed cuisines" is the difference between a broken feature and a working one. Precision in language matters as much in prompts as it does in code.
I also got hands-on experience with Lambda Function URLs as a lightweight alternative to API Gateway which is a simpler set up for a single-endpoint use case like this. And hosting a fully functional AI-powered app on S3 static hosting reinforced just how far you can get with serverless architecture and zero infrastructure overhead.
But the most important lesson wasn't technical but rather it was about delivery.
My original vision for this app included full recipes embedded directly in each meal card: ingredients, step-by-step instructions, prep and cook time, all generated by Bedrock alongside the plan. I didn't have time to build that this weekend.
I deployed without it anyway and I'm glad I did.
A working meal planner without recipes is still a working meal planner. It solves the planning problem right now. The recipe feature is a clear, well-scoped next iteration and not a reason to delay delivery. There's a real difference between an incomplete idea and an incomplete feature set. This app is complete for what it does today. Version 2 will go deeper.
Sometimes deploying something "good" is better than waiting to deploy something "perfect".
What's Next Embedded recipes — step-by-step instructions and prep/cook time on each meal card via on-demand Bedrock call Serving size scaling — grocery list quantities adjusted for household size Export to PDF — printable weekly plan and grocery list Saved plans — browse previously generated plans from DynamoDB
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article