AWS Builder Center

Revisiting My AI Meeting Prep Assistant: From Weekend MVP to a More Usable AWS Builder Showcase

A practical AWS serverless GenAI project revisited for the Builder Showcase, with usability improvements including sample input, copy-to-clipboard, and Markdown download powered by Amazon Bedrock and Nova Micro.

Earlier this summer, I built an AI Meeting Prep Assistant as part of an AWS Builder Center Weekend Challenge. The original idea was intentionally simple: take the meeting context and rough notes that I would normally review manually and turn them into a concise meeting preparation guide.
The application generates a meeting summary, key talking points, useful questions to ask, and an action checklist. My goal was to automate a small but repetitive productivity task while keeping the architecture simple enough to design, build, deploy, and document within a weekend.
For the AWS Builder Showcase, I decided not to replace the original project with something completely new. Instead, I returned to the application and asked a more practical question:
What would make this MVP easier to use in a real workflow?

What I Built

The application has a React frontend hosted using AWS Amplify Hosting.
Users provide three pieces of information:
  • a meeting title,
  • the context or purpose of the meeting,
  • and notes, email text, project updates, or other relevant background information.
The frontend sends the content through Amazon API Gateway to an AWS Lambda function.
The Lambda function builds the prompt and invokes Amazon Bedrock using Amazon Nova Micro. The generated response is returned through the API and displayed in the React application as structured Markdown.
The architecture remains deliberately serverless:
React → AWS Amplify → Amazon API Gateway → AWS Lambda → Amazon Bedrock / Amazon Nova Micro
I kept this architecture for the showcase because one of the lessons from the original weekend build was that additional complexity does not automatically make an application better.

What I Improved

For this showcase version, I focused primarily on usability.
The first addition is a Load Sample feature. A visitor can now populate the form with a realistic AWS architecture review meeting in one click. This makes the live demo easier to explore because users do not need to invent meeting content before seeing how the application works.
The second improvement is Copy Result. Once the meeting preparation has been generated, the complete Markdown response can be copied directly to the clipboard. The Markdown structure preserves headings, bullet points, bold text, and checklist items.
The third improvement is Download Markdown. The generated meeting preparation can now be downloaded as a  .md  file using a filename derived from the meeting title. This makes the result portable and easy to reuse in Markdown-compatible tools. During testing, I downloaded an architecture review and opened it in Notion, where the headings, bullet points, and action checklist were preserved cleanly.
These are relatively small features technically, but they significantly improve the workflow. The original MVP generated useful information; the showcase version makes that information easier to test, copy, save, and reuse.

What Broke Along the Way

The original build was also a useful reminder that serverless does not mean configuration-free.
During development, I had to work through API Gateway request behavior, CORS configuration, endpoint details, Lambda permissions, and the connection between Lambda and Amazon Bedrock. I also encountered the familiar distinction between opening an API URL directly in a browser using GET and invoking the actual application route using POST.
On the frontend, Markdown rendering also needed to be handled properly so that generated headings, lists, and checklists appeared as structured content rather than raw text.
During this showcase update, one surprisingly visible issue came from adding more buttons. The original layout had been designed around only Generate and Clear. Adding Load Sample, Copy Result, and Download Markdown meant revisiting spacing and responsive behavior. It was another reminder that even small product improvements can expose assumptions in the original UI.

What I Learned

The biggest lesson from revisiting the project is that improving a product does not always require adding another AWS service.
The backend architecture already solved the core problem well. Instead of adding a database, authentication, RAG, or additional AI components, I concentrated on the path between generated output and what a user might actually want to do with it next.
That resulted in a simpler and more useful second version.
This project also reinforced why I enjoy serverless development on AWS. Amplify, API Gateway, Lambda, Bedrock, and Nova Micro allowed me to move from a small productivity idea to a deployed generative AI application without managing servers.

Try the Application

I enjoyed returning to an earlier weekend project and giving it the finishing touches that were easy to postpone during the initial build.
Thanks to the AWS Builder Center community for another reason to keep building, learning, and sharing.
Inspired by: @deekshi — the GenAI image editor project showed a practical way to combine Amazon Bedrock with a serverless AWS architecture.
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