Maestro — an AI-agent platform
The ground autonomous engineers run on: a Rust control plane provisioning isolated agent sessions on GKE.
Maestro is Toptal’s internal AI-agent platform: developer sessions running as isolated containers on Google Kubernetes Engine, provisioned by a Rust/Axum control plane as GKE StatefulSets. I own the cloud infrastructure — Terraform and Helm across Cloud SQL, Redis, and Workload Identity — and the OIDC-based federated auth that eliminated static secrets.
I led the integration that lets autonomous coding agents (Claude Code, Codex, Gemini) run concurrently: a pluggable executor framework with per-agent context delivery and native Jira, Confluence, and BigQuery connectors, turning ad-hoc prompting into a governed, repeatable engineering workflow.
Holding it steady: a 20+ stage container lifecycle framework with automatic failure recovery, 80%+ test coverage across an 18-crate Rust codebase, feature-flagged rollouts, and CI guardrails — schema-drift detection, linting, coverage gates.