Yi Zhang

Designed and built with Rust and TimescaleDB, a platform that ingests openly available aeronautical, flight, and weather sources into a real-time dashboard. A live aggregation of flight plans, aircraft movements, traffic prediction, weather, and more to come.

Maintained with DevOps best practices: infrastructure as code on AWS CloudFormation, CI/CD through GitHub Actions, Grafana observability. Health-gated on-demand deployments with merely seconds of downtime. Lead time from merge to production is same-day. A warm rollback path is part of every risky change.

• 80K+ messages per minute
• Billions of messages processed and counting

Built with an AI team

• BUILD
• REVIEW
• OPS
• RESEARCH

Skywheel is my one-person project, but it is not built by myself alone. I direct a team of highly specialized LLM agents: a build lane that writes the code, an independent review lane that critiques and red-teams every change, an operations lane that deploys and monitors production, and research agents that scout ahead. The team runs on written structure, the way a good human team does: standing rules that bind every agent, from data handling to what must never be touched without sign-off; skills distilled from past incidents, so each lesson is learned once; decision records that keep settled questions settled; and issue-first planning, where a design is argued and reviewed before the first line of code. I collaborate closely with each and every agent, and every merge and every deploy passes through a human approval gate - the agents make the mechanical half abundant, and the gate is where the human half does its work: seeing more broadly, understanding more deeply, staying empathic about what we build and who it is for.

This structure is why a solo side project carries production discipline - a process that usually takes a team, because it is one.

See it live →