Hello fellow readers and amazing builders!
Every time I start an article, I am filled with this doubt whether it will actually contain a useful information. It is a valid worry, because there is so much excellent content that my efforts to put few words together seems so imbecile that I am about to quit after every sentence.
I can almost hear you saying Well, what pushed you to do it now? Excellent question!
It all started with a school project, my daughter's school project and a national competition. She was all buzzing and excited, working after school and weekends on posters, graphs and presentation. She and her team were almost ready, just finishing touches here and there when she popped a question What if we need to showcase? I heard and I knew.
It was a question resonating with every day reality.
You need proof of your idea, if you dont want it to vanish and die the second you say it out loud. To prove your concept you need a working prototype. A prototype that is light and cost effective.
The example, which I would love to share with you, and the architecture design are tailored to the project in question, but the pattern is widely and highly applicable.
The purpose of the application is to help people manage a healthy everyday routine, including the convenient use of the app via mobile device.
And there we go...question to answer...
How would the architecture look like?
First, and most importantly, this is a student project, so no budget. Same rule applies when we have PoC. We are always on a tight budget, and the rule of thumb is cost-effective. So, how do you create an application using cutting-edge technology walking on the sharpest edge of a tiny budget?
I can hear you screaming: With AWS serverless... (with an echo on less...less...).
True. So lets look at a diagram.
Very light, serverless, cost-effective.
So, here's what we have:
A single synchronous Lambda performs the entire flow end to end:
S3 ObjectCreated event triggers the Lambda.
The Lambda calls Rekognition DetectLabels on the uploaded image.
It filters the labels down to specific items (same confidence threshold + allow-list logic used in production).
It calls Bedrock InvokeModel with the objects detected list.
It stores the resulting text in DynamoDB (and/or returns it directly).
No SQS, no DLQ, no separate generation Lambda, and no retrieval API or status polling are needed. The goal is to validate the core question — "can a photo with specific objects produce a usable text for the user?" — with the fewest moving parts, giving the fastest path to build, deploy, and debug. DynamoDB is kept in the PoC because it is cheap and useful as a system of record from day one.
The cost is kept in the Free tier, simply, because is a prototype - a proof of concept. For Bedrock, a choice of cheap and light models (like Nova Micro or Nova Light) keeps the cost at minimum. One must is the AWS Budget billing alarm as a safety net.
The design is extendable, enhanceable, serverless (did I mention that already) and cheap. You can plug-in components as you go. The architecture and choice of AWS services ticks all of the requirements for a light prototype and showcase. It covers the purpose exactly.
Why do I love it so much?
For several reasons:
First (and I think most important for me), I get to be part of my daughter's school and social life.
And the rest, the project is applicable to so many situations, even kept in an abstract and high level, you can see the benefit of using these particular components. You can just grab the design and tailor it to your own idea with minimal tweaks and customization.
I hope you have enjoyed reading the article.
Please leave feedback.
Fingers crossed that the content would be useful and that it would really help someone.
Thank you!
Top comments (0)