Developer, Speaker, Tinkerer — serverless, AI agents, and whatever else is on my mind.
A fresh MicroVM took ~26 seconds from run-hook to a mounted home directory. Getting it to a few seconds taught me that on snapshot-booted compute, cold start is mostly first-touch disk I/O, and the platform's /validate hook is the fix.
A walkthrough of building a browser-based Claude Code shell that runs in an AWS Lambda MicroVM with a persistent S3 Files home. Signed in with Cognito, deployed with SAM.
Computer vision pipelines are scale problems disguised as orchestration problems. You can solve both with a single function.
Deploy a standard FastAPI app to AWS Lambda serverlessly in two commands using Lambda Web Adapter. No Docker. No handler code. No code changes.
MCP's most popular transport has no authentication. AWS serverless removes the conditions that make that dangerous.
Lambda cold starts in 2026 are 200-400ms for Python, sub-100ms for Go and Rust, and 90-140ms for Java with SnapStart. The data says the argument is over.
Lambda Managed Instances let multiple invocations share the same CPU while waiting on network calls. Here's a working demo, the cost math, and when LMI actually saves you money.
How I built a Telegram bot with Amazon Bedrock AgentCore and Claude that lets a non-developer iterate on a production PWA using plain English.
S3 Files mounts your S3 bucket as a local file system in Lambda. I built AI code review agents that share a workspace through it, orchestrated by a durable function.
A little about me, what I work on, and what you'll find on this blog.