Real systems running in my own AWS account. I pay the bill, I fix what breaks, and I learn something every time.
BoardRoom is a fully autonomous AI market research system. Every day, 17 specialized Claude agents running through Amazon Bedrock scan around 1,500 US stocks plus crypto. Screeners, researchers, a portfolio manager, and risk and macro seats each do their part, then a senior agent layer pulls everything together into briefings delivered straight to Slack. The system keeps a running journal of positions and trades, learns from its own track record by feeding past performance back into agent context, and includes a supervisor chat agent that carries out portfolio actions from plain English, with every action logged.
how it runs: It lives on an EC2 Graviton instance running Ubuntu Server 24.04 and needs zero manual intervention. The AI layer authenticates to Bedrock through an IAM instance role, so there are no static credentials anywhere. Caddy handles automatic HTTPS at the custom domain, AWS Budgets and CloudWatch keep costs in check, and GitHub Actions runs tests and a Docker build on every push.
A stock research tool I built with Flask and Python. It serves live market data, pulls in news, and generates AI summaries behind a REST style backend. It was my first real project connecting code to the cloud, and it taught me plenty about S3, CloudFront, and CI/CD along the way.
The site you're reading right now. It's hosted on S3 behind CloudFront, and every push to GitHub deploys automatically with the cache invalidated on the way. The Ask my AI assistant in the corner is a cost guarded AWS Lambda calling Amazon Bedrock, with preset questions, response caching, and hard token caps so it stays honest and cheap.