Vision & What the app does
Dealing with messy, cluttered messages every day across email and chat apps is annoying. Important tasks end up getting lost in the middle of long conversations, it is hard to figure out what actually needs to be done first, and writing back to everyone takes up way too much time.
TaskTriage AI Agent is a simple app I built to deal with that problem. You just give it raw text, and it automatically figures out how urgent the message is, pulls out a quick list of action items, and writes a ready-to-use reply draft for you. It turns messy text into clear steps so you can respond quickly and get back to work.
How I Built It
I kept things simple by using Python and AWS serverless tools.
Development Process & Key Steps:
1 Choosing the Model: I picked Amazon Nova Lite in Bedrock because it is fast, cheap to run, and does a great job understanding user requests.
2 Setting Up the Prompt: I wrote system instructions in Python to force the model to give back the exact details I needed: Urgency Rank (rated as Low, Medium, High, or Critical), Category (marked as Work, Personal, or Admin), Action Items (a list of tasks to complete), and a Suggested Reply (a short, helpful response draft).
3 Connecting the Backend: I wrote a Python script inside an AWS Lambda function using boto3 to send the text input over to Bedrock and grab the response.
Challenges I Ran Into
The main issue I ran into was setting up the right permissions in IAM so Lambda was allowed to call Bedrock. I also had to tweak my prompt a few times because the model kept adding unnecessary conversational filler instead of just giving me the formatted text I asked for.
AWS Services Used
Here are the AWS services I relied on for this project:
Amazon Bedrock: Handles the main text processing, urgency scoring, and reply writing.
AWS Lambda: Runs the Python code whenever a request comes in so I do not have to manage any servers.
AWS IAM: Sets up the security role so Lambda can safely talk to Bedrock.
How It Triggers
The app gets triggered by passing a JSON payload with raw text directly to the Lambda function. Lambda runs the Python handler, sends the text to Bedrock, and returns a status 200 response with the output.
What I Learned
Building this gave me a lot of good practice with AWS serverless tools. I got better at configuring IAM roles for service-to-service access, calling Bedrock models inside Python scripts using boto3, and designing prompts that return clean, consistent data without extra fluff.
Link to Repo
Loading article