Side projects are how I place bets on where computing is going. A bet is a thesis held until the future proves it right or wrong; the workshop below it is everything built along the way.
The bet:The next runtime for AI isn't just the datacenter. It's your browser and your own device.
An open-ended exploration, not a single project. The first artifact is a CrewAI multi-agent platform (Gemini 2.0 + GPT-4o) that crawls career portals with Playwright, acts through MCP, and hands low-confidence decisions back to a human, completing 85% of tasks. More artifacts in progress.
Built to answer one question: do I actually understand the internet? A working miniature one, with OSPF and iBGP/eBGP routing via Quagga, policy-based routing, and a custom TCP implementation with sliding windows, retransmission, and BBR congestion control.
A wager that capsule networks could beat U-Net at medical segmentation with a fraction of the parameters. It paid off: 54.21% fewer trainable parameters for sub-retinal fluid segmentation, published in Medical & Biological Engineering & Computing (Springer).
Led a team of seven through shipping a real game: a 2D space shooter in Unity/C# with custom gravity mechanics and adaptive difficulty, instrumented with Firebase, playable in the browser.