"Weekend Showcase Challenge - [š§ Meeting Memory: Turning Conversations Into Action]"
A meeting ending should not mean the important information disappears with it. Throughout the summer, I experimented with different ways to build useful AI applications. For the Builder Showcase, I decided to return to one of my strongest builds, Meeting Memory, and take it further instead of starting something completely new.
Meeting Memory is an AI-powered post-meeting assistant that transforms meeting recordings or transcripts into structured, actionable information. The idea came from a simple problem: after a meeting, the conversation is over, but the actual work has only just begun. Someone has to remember the decisions, identify action items, find the owners, track deadlines, and send follow-up messages.
I wanted AI to handle that gap.
š From Conversation to Action
Meeting Memory takes meeting content and converts it into information people can immediately use. It can generate a meeting title and summary, key discussion points, action items, task owners, deadlines, follow-up email drafts, and Slack/WhatsApp-style messages.
The core workflow is:
Meeting Content ā AI Processing ā Structured Insights ā Follow-up Actions
The important part is that Meeting Memory is not designed to simply summarize a meeting. A summary tells you what happened; an actionable assistant helps you understand what happens next.
That became the main focus of my refinement for this challenge.
šØ What I Refined
The original version was built to validate the concept. For the Showcase, I revisited the project with a different question:
How can I make the output genuinely useful after the meeting ends?
I improved the way information is structured and presented so users can quickly identify decisions, tasks, owners, and deadlines instead of reading through a large block of AI-generated text.
The project is built with Next.js, React, TypeScript, Tailwind CSS, and Framer Motion, with an AI-powered processing layer and structured prompting at its core.
I also wanted to take the project beyond a simple prototype and bring a real cloud component into the architecture.
āļø AWS Deployment
For the Builder Showcase, I deployed the Meeting Memory backend using AWS Lambda, turning the core processing workflow into a serverless cloud function.
Lambda fits Meeting Memory naturally because the backend only needs to process requests when a user submits meeting content. Instead of maintaining a traditional server continuously, the processing can run when required and scale with usage.
This deployment was an important step for me because it changed the project from simply being an AI application running as a prototype into something with a cloud-based backend architecture.
It also gives Meeting Memory a foundation for future AWS integrations such as Amazon S3 for meeting recordings, Amazon Transcribe for converting audio into transcripts, Amazon Bedrock or Amazon Nova for AWS-native AI processing, and Amazon DynamoDB for storing meetings and action items.
š§© What Broke Along the Way
The hardest part was not getting an AI response. It was getting an AI response that an application could reliably use.
Early outputs could mix summaries, tasks, owners, and deadlines together. That made the interface less predictable. I had to rethink the prompts and focus on structured extraction rather than simply asking the model to summarize the conversation.
Another challenge was deciding what actually matters in a meeting. Not every sentence should become an action item.
I therefore designed the workflow around three questions:
What needs to be done? Who owns it? When is it due?
Deployment also introduced challenges around environment variables, API configuration, request handling, Lambda setup, and connecting the application components correctly.
These problems reminded me that building an AI product is very different from simply making an AI demo.
š” What I Learned
This project taught me that the value of AI comes from the workflow built around the model.
Good prompting matters. Structured outputs matter. User experience matters. Cloud architecture matters. Most importantly, the AI needs to solve a real problem rather than simply generate impressive text.
I also learned that refining an existing project can be more valuable than constantly starting new ones. The first version gave me the foundation; this challenge gave me the opportunity to question my decisions, improve the experience, and take the idea closer to something I would actually want to use.
š± What's Next?
I see Meeting Memory becoming more than a post-meeting summarizer.
A future version could accept a recording directly, automatically transcribe it, identify decisions and action items, assign responsibilities, store them in the cloud, and help users track what happens afterward.
The long-term vision is simple:
Don't just remember the meeting. Remember what needs to happen because of it.
š Try Meeting Memory
š» GitHub Repository: Ā https://github.com/anand8795/Meeting-Memory-/tree/main/Meeting-Memory-mainĀ
āļø AWS: Deployed using AWS Lambda
This Builder Showcase is a chance to look back at everything I built this summer and choose one project worth pushing further. I chose Meeting Memory because it represents the kind of builder I want to become: someone who doesn't stop when the prototype works, but keeps asking how the idea can become more useful, reliable, and real.
A special thanks to Adtiya for inspiring me throughout this summer of building.
Build ā Break ā Learn ā Refine ā Ship. š
#application #AWS #AWSLambda #GenerativeAI #AgenticAI #AI #BuildInPublic #AWSBuilderCenter #agents #CloudComputing
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article