AWS Builder Center
Weekend Annoying Task Challenge: NotesToTasks

Weekend Annoying Task Challenge: NotesToTasks

A single AWS Lambda app that turns messy meeting notes into structured action items using Amazon Bedrock (Nova Lite), emails a summary via SES, and auto-creates matching tasks in Notion. No API Gateway, no database - one function, fully serverless, Free Tier.

Weekend Annoying Task Challenge: NotesToTasks

Tag: #productivity

Vision & What the App Does

Every meeting I sit in ends with a page of scrawled, half-finished notes and a vague sense that something needs to happen next but figuring out exactly what, who owns it, and by when usually takes another 15 minutes of re-reading and mental reconstruction after the meeting has already ended. That's the annoying task NotesToTasks kills.
You paste your raw meeting notes, however messy into a single-page app. It reads through them, pulls out every actionable item, assigns an owner where one is implied, and attaches a deadline where one is mentioned. Then, without you lifting another finger, it emails you a clean summary and creates a matching task for each item directly in a Notion database, so the follow-through lives where my team already tracks work instead of getting lost in a notes app.

How I Built It

I started by deciding the whole thing should live in one Lambda function: no API Gateway, no separate frontend hosting, no database. A Lambda Function URL serves the HTML/JS UI on  GET  and handles the extraction logic on  POST , which kept the architecture (and the deployment) genuinely simple for a weekend build.
The core extraction step uses Amazon Bedrock with the Nova Lite model via the  converse  API. I prompted it to return only a JSON array of  {task, owner, deadline}  objects explicitly no prose, no markdown. In practice, Nova Lite still occasionally wrapped its output in  ```json  fences despite being told not to, so I added a defensive regex strip before parsing. That was the single trickiest part of the build LLM output is usually almost structured, and "almost" breaks  json.loads() .
Once I have the structured items, two things happen in parallel: Amazon SES sends a plain-text summary email, and the app calls the Notion API directly (via  urllib , so no extra dependencies to package) to create one page per action item in a shared database. I made both steps fail gracefully and independently — if Notion isn't configured, the app still emails you; if email isn't configured, Notion still gets your tasks. Each step reports its own status back to the UI so you always know what actually happened.
I tested the extraction and routing logic locally by mocking the Bedrock and SES clients, which let me catch the JSON-fencing edge case before ever touching a real AWS deployment.

AWS Services Used / Architecture Overview

  • AWS Lambda (Function URL) — single entry point; serves the UI and processes requests. No API Gateway needed.
  • Amazon Bedrock (Nova Lite) — extracts structured action items from unstructured meeting notes.
  • Amazon SES — emails the summary.
  • Notion API — external integration that creates the actual tasks.
Browser --> Lambda Function URL --> Bedrock (extract)
--> SES (email summary)
--> Notion API (create tasks)

The app is triggered synchronously by the user pasting notes and clicking "Extract Action Items" — no scheduler or queue needed for this workflow, which kept the whole thing deployable with a single  sam deploy --guided .

What I Learned

I went in expecting the AWS side to be the hard part and the LLM prompting to be trivial — it was the opposite. Getting Bedrock to reliably return clean, parseable JSON took more iteration than wiring up SES or the Notion API combined, which was a good reminder to always defensively parse LLM output rather than trust the prompt alone. I also hadn't used Bedrock's  converse  API before (versus the older  invoke_model  format) and found it noticeably cleaner for structured system/user prompting.

Link to App or Repo

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