AWS Builder Center
Weekend Creative Agent Challenge: TheWalkingGhost - My Personal AI Assistant

Weekend Creative Agent Challenge: TheWalkingGhost - My Personal AI Assistant

TheWalkingGhost is my personal AI assistant built to solve my chaos of juggling cloud projects, biotech work, events, and academic deadlines. It syncs Blackboard deadlines automatically, manages my calendar through natural language, drafts emails when I am busy, sends WhatsApp reminders for critical deadlines, and provides daily morning briefings. Built with Streamlit, LangGraph, Groq API, and Supabase. Deployed on AWS ECS Fargate with Docker and ECR. #agents

Vision

TheWalkingGhost was built for me to solve my specific problems as someone constantly juggling cloud projects, biotech work, events, and academic deadlines. I needed an assistant that could keep up without opening five different apps every morning.
The problem was the friction of pulling information together from tasks, calendar, emails, and Blackboard. I was spending more time organizing than doing. TheWalkingGhost reminds me when I forget tasks, syncs Blackboard deadlines automatically, drafts emails when I am too busy, and sends WhatsApp notifications for urgent items. This agent is my second brain.

What It Does

TheWalkingGhost provides morning briefings with daily summaries of what is due, overdue, and coming up in plain English. It syncs Blackboard course deadlines automatically into my task list so I never miss assignments. I can schedule meetings or check availability through natural language. When I am in the middle of other work, I dictate emails and the agent drafts them. Critical deadlines trigger WhatsApp notifications on my phone. I can speak commands instead of typing when my hands are busy.

How I Built It

The architecture flows from browser or voice input through Streamlit UI to the LangGraph agent, which connects to Supabase for the database, Outlook API for calendar and email, and Blackboard ICS for course synchronization.
The technology stack includes Streamlit for the frontend, LangGraph for agent orchestration, Groq API for the LLM, Supabase for the database, Microsoft Graph API, Blackboard ICS, WhatsApp API, Docker for containerization, and AWS ECR, ECS Fargate, and CloudWatch for deployment.
Key technical decisions included building a date formatter that converts raw timestamps like "2026-08-30" to "August 30, 2026" and "14:30" to "2:30 PM" so the assistant is useful at a glance. I organized tasks into overdue, today's tasks, upcoming items in the next three to seven days, and historical tasks hidden by default to keep focus on what matters now. I built separate agents for tasks, calendar, email, and notifications so each has clear boundaries and responsibilities.

AWS Deployment

I needed the agent running 24/7 without relying on my local machine, so AWS provided the infrastructure to keep TheWalkingGhost always available. The services I used were Amazon ECR for Docker image storage, Amazon ECS Fargate for serverless container orchestration, AWS IAM for execution roles, Amazon CloudWatch for centralized logging, and AWS VPC for network isolation.
The deployment flow involved building the Docker image, pushing it to Amazon ECR, registering the task definition with ECS, creating the Fargate service, configuring IAM roles, and monitoring with CloudWatch.

What I Learned

Building for yourself first is important because the best tools solve your own problems. TheWalkingGhost started from personal frustration with forgetting deadlines, missing notifications, and wasting time organizing.
This challenge pushed me to learn AWS services I had never used before. Amazon ECR and pushing Docker images to a container registry was completely new to me. Amazon ECS Fargate and deploying containerized applications to AWS was a steep learning curve. I had to understand IAM roles, policies, and trust relationships. CloudWatch logging saved me when things failed during debugging.
The last day of the challenge was intense. I had to dockerize everything, push to ECR, configure ECS, fix IAM role errors, and finally see my application running with a public IP. Before this challenge, I was not using Docker much, but now I use it for everything.
Before this challenge, I worked primarily in VS Code and Streamlit locally. AWS Console was something I used occasionally. This challenge forced me to become comfortable navigating the console, configuring services, and debugging deployments directly in AWS.
Docker changed my workflow completely. Containerizing the application and deploying consistently across environments eliminated the problem of it working on my machine but not in production.
Small UX details matter. Saying "Due tomorrow at 9:59 PM" is immediately useful, while raw timestamps make users do interpretation work. Formatting is the difference between a helpful assistant and a data dump.
The most valuable part of an agent system is not the model itself but the boundaries around it. Clear tools, stable data contracts, and predictable fallbacks matter more than prompt engineering.

Challenges and Solutions

Making time human-readable was solved by building a comprehensive date formatter. Blackboard sync errors were solved by using the Supabase service role key to bypass RLS. Token persistence in containers was solved by implementing an MSAL token cache with volume mounts. ECS IAM role configuration was solved by creating a dedicated role with proper permissions. Learning AWS services was solved through hands-on trial and error and researching each service.

Live Demo

Live Application:  http://35.91.158.190:8501 
Built for the AWS Weekend Creative Agent Challenge, turning a creative app into an always-on agent.
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