
Weekend Showcase Challenge: ContextKeeper AI
I revisited ContextKeeper AI for the AWS Builder Showcase, refining my serverless second brain that turns text, images, and voice notes into organized tasks, ideas, and searchable context using AWS event-driven services and AI.
Vision & What It Does
Over the summer, I built several projects through the AWS Builder Center challenges, but ContextKeeper AI was one I wanted to revisit.
The idea came from a problem I run into often: information arrives when you are not ready to organize it.
You have an idea while working. Someone recommends something you want to remember. You take a photo of a whiteboard. You record a voice note. You think of something important and tell yourself you'll deal with it later.
Usually, that information ends up scattered across notes apps, screenshots, voice recordings, and messaging apps.
I built ContextKeeper AI as a place to capture that information naturally and let AI help turn it into something useful.
You can give ContextKeeper text, images, or audio. Depending on what you provide, the application can extract text, transcribe audio, analyze the content, identify tasks and ideas, and organize the information into structured context.
The goal is not just to store notes. I wanted the application to help answer the question: What should I do with this information now?
For example, instead of manually turning a messy brain dump into a list of tasks, ContextKeeper can identify actionable items and organize them. Instead of searching through screenshots trying to remember where you saw something, you can use natural language search to find the context later.
It also includes a Procrastination Buster workflow. Tasks that have been sitting untouched can be identified by a scheduled background process and broken into smaller, more approachable steps.
For the Builder Showcase, I went back to ContextKeeper AI and focused on making the project more showcase-ready.
One of the biggest improvements was the user experience. Previously, a visitor had to create an account before they could experience what the application actually does. That created unnecessary friction, especially for someone who just wanted to see the project in action.
The updated experience makes the core demo accessible immediately. Visitors can try ContextKeeper without going through authentication first, while accounts are still available for users who want to save and return to their own workspace.
The goal was simple: let people experience the product before asking them to sign up.
How I Built It
ContextKeeper AI uses a serverless architecture built primarily on AWS.
The frontend is built with Next.js and deployed as a static application using Amazon S3 and CloudFront. I also configured the application as a Progressive Web App to give it a more app-like experience and support offline capabilities.
The backend uses Amazon API Gateway and AWS Lambda for the main application APIs and processing logic.
For persistent application data, I use Amazon DynamoDB to store captured items, tasks, user data, and processed context. Amazon S3 handles uploaded media such as audio files and images.
The interesting part of the application is what happens after something is captured.
For text, the backend can send the content through an AI processing workflow to extract useful information such as summaries, tasks, ideas, and structured context.
For voice notes, Amazon Transcribe converts the audio into text before the content continues through the processing pipeline.
For images and documents, Amazon Textract extracts text that can then be analyzed alongside regular text input.
Some of these workflows can take longer than a normal API request. Rather than making the user wait for everything to finish, I used AWS Step Functions to coordinate the longer-running processing pipeline.
The API can accept the capture request and return quickly while Step Functions orchestrates the remaining work in the background. Depending on the type of content, the workflow can handle transcription or text extraction, AI processing, structured task extraction, embedding generation, and updating the stored item.
I also use Amazon EventBridge for scheduled workflows.
One of these workflows powers the Procrastination Buster. A scheduled process checks for tasks that have been sitting for too long and triggers an AI-powered workflow that breaks them into smaller steps.
This was one of the parts I enjoyed building because it moves ContextKeeper beyond simply reacting to user input. The application can also revisit existing information and try to make it more actionable.
For AI processing, I use OpenAI for tasks such as extracting structured information, generating summaries, and helping break down larger tasks.
AWS Services Used / Architecture Overview
ContextKeeper AI uses the following AWS services:
- Amazon S3 — Hosts the frontend and stores uploaded media such as images and audio.
- Amazon CloudFront — Distributes the frontend globally.
- Amazon API Gateway — Provides the HTTP API between the frontend and backend.
- AWS Lambda — Handles API requests, application logic, and background processing.
- Amazon DynamoDB — Stores users, captured items, tasks, and processed context.
- AWS Step Functions — Orchestrates asynchronous processing workflows.
- Amazon Transcribe — Converts uploaded voice notes into text.
- Amazon Textract — Extracts text from images and documents.
- Amazon EventBridge — Runs scheduled workflows, including the Procrastination Buster.
- Amazon Cognito — Handles authentication for users who want to save and manage their own workspace.
The architecture separates the fast user-facing request from longer-running processing. This means someone can capture information without waiting for transcription, OCR, AI analysis, and other processing steps to complete inside a single HTTP request.

How the Architecture Works
When a user captures something, the request first goes through API Gateway to a Lambda function.
Depending on the type of input, the content is stored and passed into a Step Functions workflow.
An audio capture can go through Amazon Transcribe. An image or document can go through Amazon Textract. The extracted text is then passed through the AI processing layer, which identifies useful information such as summaries, tasks, ideas, and other structured context.
The processed result is stored in DynamoDB and becomes available to the application.
Separately, EventBridge can trigger scheduled workflows without requiring the user to open the application. This is how the Procrastination Buster can revisit old tasks and generate smaller steps to help make them easier to start.
What I Learned Across the Summer
One of the biggest things I learned from the summer challenges was how much easier it is to build when you stop trying to make every project perfect before shipping it.
Each challenge pushed me to take an idea, make decisions quickly, build something real, deploy it, and then explain how it works.
Across projects like BuilderOS AI, Orbit, ContextKeeper AI, and QuestForge, I also spent more time working with serverless architectures and event-driven systems.
I became much more comfortable thinking in terms of triggers, background workflows, scheduled jobs, and managed services instead of building everything around a traditional always-running backend.
ContextKeeper also reinforced an important lesson about user experience.
A project can have a lot happening behind the scenes, but if someone cannot quickly understand what it does or try it without unnecessary friction, the technology underneath does not matter much to the person seeing it for the first time.
That is one of the main reasons I chose ContextKeeper for this Showcase. Going back to the project gave me the opportunity to improve not just the infrastructure, but the way someone actually experiences it.
The summer challenges also showed me the value of iteration.
The first version of a project does not need to be the final version. Sometimes the best next step is simply to revisit something you already built, look at what was rushed or incomplete, and make it better.
Try ContextKeeper AI
You can try the live application here:
You can also explore the source code:
A builder who inspired me throughout the summer was @LewisSawe. I enjoyed seeing how he approached the challenges and turned creative ideas into projects that were both interesting and technically thoughtful.
ContextKeeper AI started as a way to stop useful information from disappearing into the noise. Bringing it back for the Builder Showcase gave me the chance to take it further and focus on something just as important as the architecture behind it: making it easier for people to actually experience what I built.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article